GPT-5.6 Sol

Closed weights · Large · Price proxy

Token price: $4 input / $20 output per million. Blended 3:1: $8/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges. Published $4/$20 rate is promotional through at least November 21, 2026.

• OpenAI GPT-5.6 Sol API model documentation

Checked 2026-10-10 · high confidence

GPT-6 Astra

Closed weights · Large · Price proxy

Token price: $10 input / $50 output per million. Blended 3:1: $20/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-6 Astra API model documentation

Checked 2026-10-10 · high confidence

Kimi K3

Open weights · Large · Published size · 2,800B total · 104B active

Token price: $3 input / $15 output per million. Blended 3:1: $6/M.

Released Kimi K3 License weights verified in the public repository (96 safetensors shards). Total parameters determine category; active parameters describe inference. Launch API uncached rates shown.

• Moonshot Kimi K3 model card
• Moonshot Kimi K3 announcement

Checked 2026-10-10 · high confidence

Claude Opus 5

Closed weights · Large · Price proxy

Token price: $5 input / $25 output per million. Blended 3:1: $10/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

GPT-5.6 Sol

Closed weights · Large · Price proxy

Token price: $4 input / $20 output per million. Blended 3:1: $8/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges. Published $4/$20 rate is promotional through at least November 21, 2026.

• OpenAI GPT-5.6 Sol API model documentation

Checked 2026-10-10 · high confidence

GPT-5.5 Pro

Closed weights · Large · Price proxy

Token price: $30 input / $180 output per million. Blended 3:1: $67.5/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.5 Pro API model documentation

Checked 2026-10-10 · high confidence

GPT-5.4 Pro

Closed weights · Large · Price proxy

Token price: $30 input / $180 output per million. Blended 3:1: $67.5/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 Pro API model documentation

Checked 2026-10-10 · high confidence

MiroThinker-H1 model stack

Uncertain weights · Large · Estimated

Provisional large: H1 extends the 235B-class MiroThinker-1.7 with local/global verification, but exact H1 models, weights and token prices are not disclosed. The public repository explicitly calls H1 proprietary; released 1.7 weights do not establish that the entire H1 stack is open.

• MiroThinker official repository
• MiroThinker-1.7 and H1 paper

Checked 2026-10-10 · low confidence

Claude Mythos 5

Closed weights · Large · Price proxy

Token price: $10 input / $50 output per million. Blended 3:1: $20/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

Claude Fable 5

Closed weights · Large · Price proxy

Token price: $10 input / $50 output per million. Blended 3:1: $20/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

Claude Mythos Preview

Closed weights · Large · Price proxy

Token price: $25 input / $125 output per million. Blended 3:1: $50/M.

Undisclosed parameters. Published participant API price; limited research-preview access, with initial usage credits. Token pricing excludes tools.

• Anthropic Project Glasswing

Checked 2026-10-10 · high confidence

Kimi K2.6

Open weights · Large · Published size · 1,000B total · 32B active

Modified-MIT weights verified in public repository (64 safetensors shards). Swarm orchestration is distinct from these model weights. Price table was absent from retrieved pricing text and direct retrieval was blocked; token rates omitted.

• Moonshot Kimi K2.6 model card
• Kimi API pricing

Checked 2026-10-10 · high confidence

Gemini 3.1 Pro

Closed weights · Large · Price proxy

Token price: $2 input / $12 output per million. Blended 3:1: $4.5/M.

Parameters are undisclosed. Gemini 3.1 Pro Preview standard text rates for prompts up to 200K tokens; larger prompts and grounding cost more.

• Google Gemini API pricing

Checked 2026-10-10 · high confidence

Claude Sonnet 5

Closed weights · Large · Price proxy

Token price: $2 input / $10 output per million. Blended 3:1: $4/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

GPT-5.5

Closed weights · Large · Price proxy

Token price: $5 input / $30 output per million. Blended 3:1: $11.25/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.5 API model documentation

Checked 2026-10-10 · high confidence

Claude Opus 4.8

Closed weights · Large · Price proxy

Token price: $5 input / $25 output per million. Blended 3:1: $10/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

AREX-2

Open weights · Small · Published size · 27B total

BAAI fine-tune of Qwen3.8-27B; 27B is the published nominal total. Publisher repository BAAI/AREX-2 is public, Apache-2.0, and contains 11 safetensors shards (27.3567B serialized parameters). The official research profile reuses this policy for page summaries; external judges only score benchmark answers.

• AREX-2 paper, model and evaluation protocol
• BAAI AREX-2 released model and checkpoint inventory
• AREX-2 research profile and summary-model configuration

Checked 2026-10-11 · high confidence

Claude Opus 4.6

Closed weights · Large · Price proxy

Token price: $5 input / $25 output per million. Blended 3:1: $10/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

DeepSeek-V4-Pro

Open weights · Large · Published size · 1,600B total · 49B active

MIT weights verified in public repository. Max is a reasoning-effort mode of this checkpoint, not a larger model. Current API pricing names later checkpoints, so it is not attributed to this April benchmark model.

• DeepSeek V4 Pro model card
• DeepSeek API model versions and pricing

Checked 2026-10-10 · high confidence

GPT-5.4

Closed weights · Large · Price proxy

Token price: $2.5 input / $15 output per million. Blended 3:1: $5.625/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 API model documentation

Checked 2026-10-10 · high confidence

Claude Opus 4.7

Closed weights · Large · Price proxy

Token price: $5 input / $25 output per million. Blended 3:1: $10/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

GLM-5.1

Open weights · Large · Published size · 744B total · 40B active

Token price: $1.4 input / $4.4 output per million. Blended 3:1: $2.15/M.

MIT weights verified. Official GLM repository model table specifies 744B total / 40B active; category uses total. Published Z.ai API rates exclude search charges.

• GLM official model table
• GLM-5.1 model card
• Z.ai API pricing

Checked 2026-10-10 · high confidence

Qwen3.5-397B-A17B

Open weights · Large · Published size · 397B total · 17B active

Apache-2.0 weights verified (94 safetensors shards); published architecture gives 397B total / 17B active. No version-matched first-party token price verified.

• Qwen official model card

Checked 2026-10-10 · high confidence

Kimi K2.5

Open weights · Large · Published size · 1,000B total · 32B active

Modified-MIT weights verified in public repository (64 safetensors shards). Swarm orchestration is distinct from these model weights. Price table was absent from retrieved pricing text and direct retrieval was blocked; token rates omitted.

• Moonshot Kimi K2.5 model card
• Kimi API pricing

Checked 2026-10-10 · high confidence

GPT-5.2 Pro

Closed weights · Large · Price proxy

Token price: $21 input / $168 output per million. Blended 3:1: $57.75/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.2 Pro API model documentation

Checked 2026-10-10 · high confidence

GPT-5.3-Codex

Closed weights · Large · Price proxy

Token price: $1.75 input / $14 output per million. Blended 3:1: $4.8125/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.3-Codex API model documentation

Checked 2026-10-10 · high confidence

Seed 2.0 Pro

Closed weights · Large · Estimated

Provisional large for the proprietary flagship. Parameters and standard USD rates for the evaluated 0215 snapshot were not verified. BytePlus current pricing embeds a later 260328 snapshot; those rates are not assigned to this model.

• ByteDance Seed2 release
• BytePlus model pricing

Checked 2026-10-10 · low confidence

Inkling

Open weights · Large · Published size · 975B total · 41B active

Token price: $1 input / $4.05 output per million. Blended 3:1: $1.7625/M.

Apache-2.0 full weights verified. Publisher reports 975B total / 41B active. Rates are Tinker serverless inference beta, standard uncached prefill/sample, excluding training.

• Thinking Machines Inkling announcement
• Inkling released weights
• Tinker models and serverless pricing

Checked 2026-10-10 · high confidence

MiniMax M2.5

Open weights · Large · Published size · 229.9B total · 9.8B active

Token price: $0.3 input / $1.2 output per million. Blended 3:1: $0.525/M.

Modified-MIT weights verified. M2-series report gives 229.9B total / 9.8B active backbone. Current standard API rates shown; launch-card input pricing differs. Category uses total parameters.

• MiniMax M2.5 weights
• MiniMax M2-series technical report
• MiniMax API pricing

Checked 2026-10-10 · high confidence

Claude Sonnet 4.6

Closed weights · Large · Price proxy

Token price: $3 input / $15 output per million. Blended 3:1: $6/M.

Undisclosed parameters; size uses standard uncached token pricing. Tool charges excluded.

• Anthropic model pricing

Checked 2026-10-10 · high confidence

GLM-5

Open weights · Large · Published size · 744B total · 40B active

Token price: $1 input / $3.2 output per million. Blended 3:1: $1.55/M.

MIT weights verified. Official GLM repository model table specifies 744B total / 40B active; category uses total. Published Z.ai API rates exclude search charges.

• GLM official model table
• GLM-5 model card
• Z.ai API pricing

Checked 2026-10-10 · high confidence

Agents-A1

Open weights · Small · Published size · 35B total · 3B active

Apache-2.0 35B model weights verified (14 safetensors shards). Fine-tuned from Qwen3.5-35B-A3B; 3B active, but total parameters set the small category. No standard API token price verified.

• Agents-A1 model card
• Agents-A1 paper

Checked 2026-10-10 · high confidence

GPT-4o mini

Closed weights · Medium · Price proxy

Token price: $0.15 input / $0.6 output per million. Blended 3:1: $0.2625/M.

Standard uncached input/output pricing gives a $0.2625/M blend: Medium price proxy. InfoAgent uses this hosted model for search snippets. Parameter count is undisclosed.

• GPT-4o mini model documentation

Checked 2026-10-10 · high confidence

Agents-A1-SFT

Uncertain weights · Small · Published size · 35B total · 3B active

35B/3B SFT-stage checkpoint from the Agents-A1 paper. Its exact weights were not found in the publisher model listing; the guessed SFT repository returned HTTP 401. The released final Agents-A1 does not prove release of the SFT checkpoint. No API token price verified.

• Agents-A1 paper
• InternScience model listing

Checked 2026-10-10 · medium confidence

GPT-4o mini

Closed weights · Medium · Price proxy

Token price: $0.15 input / $0.6 output per million. Blended 3:1: $0.2625/M.

Standard uncached input/output pricing gives a $0.2625/M blend: Medium price proxy. InfoAgent uses this hosted model for search snippets. Parameter count is undisclosed.

• GPT-4o mini model documentation

Checked 2026-10-10 · high confidence

Exa Agent model stack

Uncertain weights · Large · Estimated

Provisional large: Exa describes a mixture of frontier and cost-effective models with its search/highlights tools. Exact model IDs, weights, parameters and token prices are undisclosed. Per-query service costs cannot substitute for per-token model pricing.

• Exa Agent announcement

Checked 2026-10-10 · low confidence

DeepSeek-V4-Flash

Open weights · Large · Published size · 284B total · 13B active

MIT weights verified in public repository. Max is a reasoning-effort mode of this checkpoint, not a larger model. Current API pricing names later checkpoints, so it is not attributed to this April benchmark model.

• DeepSeek V4 Flash model card
• DeepSeek API model versions and pricing

Checked 2026-10-10 · high confidence

LongCat-Flash-Thinking-2601

Open weights · Large · Published size · 560B total · 27B active

MIT weights are available. Card reports 560B total and 27B activated; Heavy Thinking is an inference mode of this checkpoint. No standard USD token price verified.

• meituan-longcat/LongCat-Flash-Thinking-2601 model card
• meituan-longcat/LongCat-Flash-Thinking-2601 weight files

Checked 2026-10-10 · high confidence

Traverse-35B

Uncertain weights · Small · Published size · 35B total · 3B active

Qwen3.5-35B-A3B-based trained search policy; 35B nominal total determines Small, with 3B active per token. The official runtime repository is public but separates policy weights from the harness. No exact evaluated checkpoint inventory was located; weights remain uncertain. The evaluated system also has a separate page-summary tool whose backbone is undisclosed.

• Traverse paper, implementation details and Table 1

Checked 2026-10-11 · medium confidence

Traverse page-summary backend (undisclosed)

Uncertain weights · Medium · Estimated

Paper Appendix F.1.2 supplies question-conditioned synthesized page answers and permits a reader-plus-LLM fallback. The public runtime delegates link_summary_tool to a deployment-specific callable and does not identify its evaluated backbone, parameters, or token rate. Medium is a provisional estimate for this bounded summarization role, consistent with other unresolved research summarizers; its actual size and whether it reuses the actor remain unknown. No parameter count is invented.

• Traverse page-summary interface and reader-plus-LLM fallback
• Traverse deployment-specific search and summary backends
• Traverse FunctionLinkSummaryTool wrapper

Checked 2026-10-11 · low confidence

Nemotron 3.5 Bolt 74B-A9B-BF16

Open weights · Medium · Published size · 74B total · 9B active

NVIDIA publishes 74B total / 9B active parameters. Full-precision BF16 checkpoint. Open weights under the OpenMDW License Agreement v1.1.

• nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB-BF16 model card

Checked 2026-10-14 · high confidence

Step-3.5-Flash

Open weights · Large · Published size · 196.81B total · 11B active

Apache-2.0 weights are available. Card specifies 196B backbone plus 0.81B head and approximately 11B active. No standard USD token price verified.

• stepfun-ai/Step-3.5-Flash model card
• stepfun-ai/Step-3.5-Flash weight files

Checked 2026-10-10 · high confidence

GLM-4.7

Open weights · Large · Published size · 358.338B total

Token price: $0.6 input / $2.2 output per million. Blended 3:1: $1/M.

MIT weight shards verified. Total is the publisher repository safetensors count, rounded to four decimals in billions; it includes stored auxiliary tensors. Active count was not established by the checked model card. Z.AI lists $0.60 input / $2.20 output per 1M tokens.

• zai-org/GLM-4.7 model card
• zai-org/GLM-4.7 weight files
• Z.AI standard USD pricing
• Publisher weight metadata

Checked 2026-10-10 · high confidence

Nemotron 3.5 Bolt 74B-A9B-24GB

Open weights · Medium · Published size · 74B total · 9B active

NVIDIA publishes 74B total / 9B active parameters. IQ2-quantized checkpoint, about 24GB. Open weights under the OpenMDW License Agreement v1.1.

• nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-24GB model card

Checked 2026-10-14 · high confidence

GPT-5.2

Closed weights · Large · Price proxy

Token price: $1.75 input / $14 output per million. Blended 3:1: $4.8125/M.

Parameter count is undisclosed. Large is a price proxy from standard uncached API text rates; it does not assert a physical parameter size. Tool fees and cached, batch, flex, and priority rates are excluded.

• OpenAI GPT-5.2 API model documentation

Checked 2026-10-10 · high confidence

Qwen3.5-122B-A10B

Open weights · Medium · Published size · 122B total · 10B active

Apache-2.0 weights are available. Uses the card’s nominal 122B total / 10B active count. No standard USD token price for this exact checkpoint verified.

• Qwen/Qwen3.5-122B-A10B model card
• Qwen/Qwen3.5-122B-A10B weight files

Checked 2026-10-10 · high confidence

MiniMax M2.1

Open weights · Large · Published size · 228.69B total

Token price: $0.3 input / $1.2 output per million. Blended 3:1: $0.525/M.

Modified-MIT weight shards verified. Total is the publisher repository safetensors count, rounded in billions. Active count was not established for this exact version by the checked card. Standard legacy-model API rates are $0.30 input / $1.20 output per 1M.

• MiniMaxAI/MiniMax-M2.1 model card
• MiniMaxAI/MiniMax-M2.1 weight files
• MiniMax standard pay-as-you-go pricing
• Publisher weight metadata

Checked 2026-10-10 · high confidence

LongSeeker-30B-SFT

Open weights · Small · Published size · 30.5B total · 3.3B active

Available MIT finetuned weights identify Qwen3-30B-A3B-Thinking-2507 as the base. Count follows that published base architecture and matches the 30.532B stored tensors. No standard token price published in the checked sources.

• PolarSeeker/LongSeeker-30B-SFT model card
• PolarSeeker/LongSeeker-30B-SFT weight files
• Qwen3 base architecture
• LongSeeker paper

Checked 2026-10-10 · high confidence

LongSeeker undisclosed Visit summarizer

Uncertain weights · Medium · Estimated

Runtime visit_web uses SUMMARY_MODEL_NAME, which is left blank in the public configuration. The evaluated helper identity, weights, parameters and price are unresolved. Medium is a low-confidence provisional category informed by the related WebSailor Visit pipeline’s 72B helper; it is not an asserted parameter count or confirmed identity. DeepSeek-V3.2 is named as a training teacher, which does not establish that it is this runtime helper.

• LongSeeker summary-model configuration and tools
• LongSeeker runtime summary client
• LongSeeker paper
• WebSailor inference instructions

Checked 2026-10-10 · low confidence

Qwen3.5-27B

Open weights · Small · Published size · 27B total · 27B active

Apache-2.0 dense weights are available; all 27B nominal parameters are active. No standard USD token price for this exact checkpoint verified.

• Qwen/Qwen3.5-27B model card
• Qwen/Qwen3.5-27B weight files

Checked 2026-10-10 · high confidence

Qwen3.5-35B-A3B

Open weights · Small · Published size · 35B total · 3B active

Apache-2.0 weights are available. Uses published 35B total / 3B active count. No standard USD token price for this exact checkpoint verified.

• Qwen/Qwen3.5-35B-A3B model card
• Qwen/Qwen3.5-35B-A3B weight files

Checked 2026-10-10 · high confidence

Kimi K2 Thinking

Open weights · Large · Published size · 1,000B total · 32B active

Modified-MIT downloadable weights verified; card reports 1T total / 32B active. Current pricing page did not expose this historical model in readable text and direct retrieval returned HTTP 403, so token prices remain unset.

• moonshotai/Kimi-K2-Thinking model card
• moonshotai/Kimi-K2-Thinking weight files

Checked 2026-10-10 · high confidence

Gemini 3 Pro

Closed weights · Large · Estimated

Provisional Large based on Google’s flagship Pro positioning. Parameters are undisclosed; current Google pricing pages have replaced this version with Gemini 3.1 Pro, so no exact-version current token rate is asserted.

• Google Gemini 3.1 Pro model card, including Gemini 3 Pro comparison
• Current Gemini pricing, checked for exact-version coverage

Checked 2026-10-10 · low confidence

MiMo-V2-Flash

Open weights · Large · Published size · 309B total · 15B active

MIT downloadable weights verified. Card states 309B total / 15B active. No standard USD token price verified; Flash does not mean Small under the parameter rule.

• XiaomiMiMo/MiMo-V2-Flash model card
• XiaomiMiMo/MiMo-V2-Flash weight files

Checked 2026-10-10 · high confidence

Parallel Task Ultra model stack

Uncertain weights · Large · Estimated

Provisional Large for the hosted high-compute research stack. The checked vendor page does not identify exact underlying weights or per-token pricing. Ultra multipliers increase task compute; they are not parameter counts. Per-query cost is not used as token pricing.

• Parallel September 2025 benchmark report

Checked 2026-10-10 · low confidence

Parallel Task Ultra model stack

Uncertain weights · Large · Estimated

Provisional Large for the hosted high-compute research stack. The checked vendor page does not identify exact underlying weights or per-token pricing. Ultra multipliers increase task compute; they are not parameter counts. Per-query cost is not used as token pricing.

• Parallel September 2025 benchmark report

Checked 2026-10-10 · low confidence

GPT-5

Closed weights · Large · Price proxy

Token price: $1.25 input / $10 output per million. Blended 3:1: $3.4375/M.

Parameter count is undisclosed. Large is a price proxy from standard uncached API text rates; it does not assert a physical parameter size. Tool fees and cached, batch, flex, and priority rates are excluded.

• OpenAI GPT-5 API model documentation

Checked 2026-10-10 · high confidence

GPT-5.4

Closed weights · Large · Price proxy

Token price: $2.5 input / $15 output per million. Blended 3:1: $5.625/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 API model documentation

Checked 2026-10-10 · high confidence

Exa Agent Medium model stack

Uncertain weights · Large · Estimated

Provisional Large: vendor describes a fusion of frontier and cost-effective language models but does not identify the Medium tier’s exact model mix or release underlying weights. No standard per-token price can be assigned to its per-query charge.

• Exa Agent launch and model-fusion description

Checked 2026-10-10 · low confidence

OpenAI original Deep Research model

Closed weights · Large · Estimated

Original 2025 BrowseComp research system. Provisional Large based on its specialized persistent-browsing research positioning. The later o3-deep-research API snapshot is dated June 26, 2025; its token rates are not assigned to the earlier February/April system without an exact checkpoint match.

• OpenAI deep research API model and dated snapshot
• BrowseComp paper

Checked 2026-10-10 · low confidence

DeepSeek-V3.2

Open weights · Large · Published size · 685.355B total

MIT downloadable weights verified. Total is publisher safetensors metadata, rounded in billions, including auxiliary tensors. V3.2 adds sparse-attention indexing to the V3.1 architecture; active count was not independently established here. Current API pricing page now describes successor models, so historical V3.2 prices remain unset.

• deepseek-ai/DeepSeek-V3.2 model card
• deepseek-ai/DeepSeek-V3.2 weight files
• DeepSeek-V3.2 technical report
• Publisher weight metadata

Checked 2026-10-10 · high confidence

DeepSeek-V3.2

Open weights · Large · Published size · 685.355B total

MIT downloadable weights verified. Total is publisher safetensors metadata, rounded in billions, including auxiliary tensors. V3.2 adds sparse-attention indexing to the V3.1 architecture; active count was not independently established here. Current API pricing page now describes successor models, so historical V3.2 prices remain unset.

• deepseek-ai/DeepSeek-V3.2 model card
• deepseek-ai/DeepSeek-V3.2 weight files
• DeepSeek-V3.2 technical report
• Publisher weight metadata

Checked 2026-10-10 · high confidence

GPT-5.4

Closed weights · Large · Price proxy

Token price: $2.5 input / $15 output per million. Blended 3:1: $5.625/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 API model documentation

Checked 2026-10-10 · high confidence

Parallel Task Ultra model stack

Uncertain weights · Large · Estimated

Provisional Large for the hosted high-compute research stack. The checked vendor page does not identify exact underlying weights or per-token pricing. Ultra multipliers increase task compute; they are not parameter counts. Per-query cost is not used as token pricing.

• Parallel September 2025 benchmark report

Checked 2026-10-10 · low confidence

o3

Closed weights · Large · Price proxy

Token price: $2 input / $8 output per million. Blended 3:1: $3.5/M.

Parameter count is undisclosed. Large is a price proxy from standard uncached API text rates; it does not assert a physical parameter size. Tool fees and cached, batch, flex, and priority rates are excluded.

• OpenAI o3 API model documentation

Checked 2026-10-10 · high confidence

Sarvam-105B

Open weights · Medium · Published size · 105B total · 10.3B active

Apache-2.0 downloadable weights verified. Published card identifies a 105B MoE with 10.3B active. No standard USD token rate verified.

• sarvamai/sarvam-105b model card
• sarvamai/sarvam-105b weight files

Checked 2026-10-10 · high confidence

Perplexity Agent Advanced historical stack

Uncertain weights · Large · Estimated

Provisional Large for the historical advanced research preset. Current documentation explicitly says dynamic presets can change their underlying model; advanced-deep-research has been renamed high. Exact June 2026 backbone is not established, so no current preset model or per-query fee is misrepresented as its token price.

• Perplexity preset configuration and versioning
• Exa June 2026 comparison

Checked 2026-10-10 · low confidence

SMTL-30B

Open weights · Small · Published size · 30.532B total

Apache-2.0 weight shards verified. Sparse Qwen3 architecture is confirmed in config.json; total comes from 30,532,122,624 published tensors. Card is otherwise empty, so exact base checkpoint and active count are not asserted. No standard token price verified.

• PersonalAILab/SMTL-30B model card
• PersonalAILab/SMTL-30B weight files
• SMTL model configuration
• SMTL technical report

Checked 2026-10-10 · high confidence

DeepSeek-V3.2

Open weights · Large · Published size · 685.355B total

MIT downloadable weights verified. Total is publisher safetensors metadata, rounded in billions, including auxiliary tensors. V3.2 adds sparse-attention indexing to the V3.1 architecture; active count was not independently established here. Current API pricing page now describes successor models, so historical V3.2 prices remain unset.

• deepseek-ai/DeepSeek-V3.2 model card
• deepseek-ai/DeepSeek-V3.2 weight files
• DeepSeek-V3.2 technical report
• Publisher weight metadata

Checked 2026-10-10 · high confidence

MiroThinker v1.0-72B

Open weights · Medium · Published size · 72B total · 72B active

MIT downloadable dense weights verified; model card identifies Qwen2.5-72B-Instruct as the base. No standard USD token price published in the checked card.

• miromind-ai/MiroThinker-v1.0-72B model card
• miromind-ai/MiroThinker-v1.0-72B weight files

Checked 2026-10-10 · high confidence

MiroThinker v1.0 undisclosed webpage summarizer

Uncertain weights · Small · Estimated

The v1.0 agent configuration enables jina_scrape_llm_summary with a separate SUMMARY_LLM_MODEL_NAME left unpinned. Its paper describes a lightweight summarizer and gives Qwen3-14B as an example, supporting a provisional Small category but not a confirmed identity or count. Current README also allows open Qwen3-14B or closed GPT-5-Nano; full-stack weight availability is therefore unresolved. Post-run grading models are excluded.

• MiroThinker v1.0 paper tool description
• MiroThinker v1.0 tool configuration
• MiroThinker runtime summary tool
• MiroThinker summary setup instructions

Checked 2026-10-10 · low confidence

Perplexity Agent Deep historical stack

Uncertain weights · Large · Estimated

Provisional Large for the historical deep research preset. Current docs rename deep-research to medium and state backbones can change; the exact June 2026 model is unresolved. This category is a low-confidence editorial estimate, not derived from the product effort name or query fee.

• Perplexity preset configuration and versioning
• Exa June 2026 comparison

Checked 2026-10-10 · low confidence

OpenSeeker-v2-30B-SFT

Open weights · Small · Published size · 30.532B total

Apache-2.0 weight shards verified; config confirms sparse Qwen3 architecture. Total from 30,532,122,624 published tensors. Exact base revision and active count are not separately asserted. No standard USD token price published in the checked sources.

• PolarSeeker/OpenSeeker-v2-30B-SFT model card
• PolarSeeker/OpenSeeker-v2-30B-SFT weight files

Checked 2026-10-10 · high confidence

OpenSeeker undisclosed Visit summarizer

Uncertain weights · Medium · Estimated

A webpage summary model is required, but evaluated identity and weights are not published in the inspected source/configuration. Medium is a low-confidence provisional category informed by the related WebSailor pipeline’s 72B summary helper; it is not an asserted parameter count or confirmed model identity. This is the runtime Visit helper; the training-only historical-context teacher is excluded.

• OpenSeeker runtime tool calls
• OpenSeeker Visit implementation
• WebSailor inference instructions

Checked 2026-10-10 · low confidence

WebAnchor-30B

Uncertain weights · Small · Published size · 30B total · 3B active

Paper trains the 30B Tongyi DeepResearch base with Anchor-GRPO. Total and active counts follow that cited 30B/3B base. No WebAnchor-specific downloadable weights or harness release was found in the paper or Hugging Face model search; an open base does not establish open finetuned weights. No standard token price verified.

• WebAnchor paper, base policies and training
• Tongyi DeepResearch base architecture

Checked 2026-10-10 · medium confidence

GLM-4.6

Open weights · Large · Published size · 356.786B total

Token price: $0.6 input / $2.2 output per million. Blended 3:1: $1/M.

MIT weight shards verified. Total is publisher safetensors count, rounded in billions, including auxiliary tensors. Active count was not established by checked exact-version card. Standard Z.AI price is $0.60 input / $2.20 output per 1M.

• zai-org/GLM-4.6 model card
• zai-org/GLM-4.6 weight files
• Z.AI standard USD pricing
• Publisher weight metadata

Checked 2026-10-10 · high confidence

Parallel Task Ultra model stack

Uncertain weights · Large · Estimated

Provisional Large for the hosted high-compute research stack. The checked vendor page does not identify exact underlying weights or per-token pricing. Ultra multipliers increase task compute; they are not parameter counts. Per-query cost is not used as token pricing.

• Parallel September 2025 benchmark report

Checked 2026-10-10 · low confidence

Grok 4 Fast

Closed weights · Medium · Price proxy

Token price: $0.2 input / $0.5 output per million. Blended 3:1: $0.275/M.

Parameter count undisclosed. Standard uncached short-context rates; the launch lists higher long-context rates. Medium is a price proxy, not a parameter estimate.

• Grok 4 Fast launch and standard pricing

Checked 2026-10-10 · high confidence

MiniMax M2

Open weights · Large · Published size · 230B total · 10B active

Published architectural total; checkpoint storage is approximately 228.7B tensors. Released modified-MIT checkpoint contains 130 safetensor shards. Active count does not determine size.

• MiniMaxAI/MiniMax-M2 model card and released files
• MiniMaxAI/MiniMax-M2 checkpoint inventory

Checked 2026-10-10 · high confidence

Exa Agent Low undisclosed model stack

Uncertain weights · Large · Estimated

Exa describes a combination of top language models and model fusion without naming this evaluated tier’s backbones. Large is a provisional frontier-model inference. No token prices or parameter counts inferred from its per-query cost; exact weights remain uncertain.

• Exa Agent launch

Checked 2026-10-10 · low confidence

Tongyi DeepResearch 30B-A3B

Open weights · Small · Published size · 30.5B total · 3.3B active

Repository publishes 30.5B total and 3.3B active; model card rounds these to 30B/3B. Exact checkpoint weights are released. A separate runtime summary model is tracked at submission level.

• Alibaba-NLP/Tongyi-DeepResearch-30B-A3B model card and released files
• Alibaba-NLP/DeepResearch repository

Checked 2026-10-10 · high confidence

Tongyi DeepResearch undisclosed summarizer

Uncertain weights · Medium · Estimated

A webpage summary model is required, but evaluated identity and weights are not published in the inspected source/configuration. Medium is a low-confidence provisional category informed by the related WebSailor pipeline’s 72B summary helper; it is not an asserted parameter count or confirmed model identity.

• Tongyi DeepResearch tool appendix
• Summary model configuration
• WebSailor inference instructions

Checked 2026-10-10 · low confidence

GLM-4.7-Flash

Open weights · Small · Published size · 30B total · 3B active

Model card describes 30B-A3B; MIT checkpoint files are publicly available. Stored tensor inventory is approximately 31.2B, without affecting the category.

• zai-org/GLM-4.7-Flash model card and released files
• zai-org/GLM-4.7-Flash checkpoint inventory

Checked 2026-10-10 · high confidence

GPT-5.4

Closed weights · Large · Price proxy

Token price: $2.5 input / $15 output per million. Blended 3:1: $5.625/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 API model documentation

Checked 2026-10-10 · high confidence

DeepSeek-V3.2-Exp

Open weights · Large · Published size · 685.4B total

Released MIT checkpoint; 685.397B is the repository safetensors inventory rounded to 0.1B, including auxiliary tensors. The card says V3.1-Terminus plus sparse attention; exact activated count is not asserted here.

• deepseek-ai/DeepSeek-V3.2-Exp model card and released files
• deepseek-ai/DeepSeek-V3.2-Exp checkpoint inventory

Checked 2026-10-10 · high confidence

GPT-5.4

Closed weights · Large · Price proxy

Token price: $2.5 input / $15 output per million. Blended 3:1: $5.625/M.

Parameters are undisclosed. Category uses published standard uncached API token prices, not an estimated parameter count. Prices exclude tools and long-context surcharges.

• OpenAI GPT-5.4 API model documentation

Checked 2026-10-10 · high confidence

AgentFounder-30B

Uncertain weights · Small · Published size · 30B total · 3B active

Post-trained Qwen3-30B-A3B variant from the paper. No exact AgentFounder evaluated checkpoint was located; released Tongyi DeepResearch is a different result and is not substituted.

• AgentFounder paper
• AgentFounder project page

Checked 2026-10-10 · medium confidence

AgentFounder undisclosed summarizer

Uncertain weights · Medium · Estimated

A webpage summary model is required, but evaluated identity and weights are not published in the inspected source/configuration. Medium is a low-confidence provisional category informed by the related WebSailor pipeline’s 72B summary helper; it is not an asserted parameter count or confirmed model identity.

• AgentFounder paper
• WebSailor inference instructions

Checked 2026-10-10 · low confidence

Sarvam-30B

Open weights · Small · Published size · 30B total · 2.4B active

Published 30B architectural class; full checkpoint inventory is approximately 32.15B. The reported 2.4B active count excludes embeddings. Apache-2.0 weights are available.

• sarvamai/sarvam-30b model card and released files
• sarvamai/sarvam-30b checkpoint inventory

Checked 2026-10-10 · high confidence

DeepMiner-32B

Uncertain weights · Small · Published size · 32B total

Paper identifies an RL-trained Qwen3-32B checkpoint with dynamic context windows and no external summarization model. No author-hosted evaluated checkpoint was located; an unrelated similarly named HF upload is not evidence of release.

• DeepMiner paper

Checked 2026-10-10 · medium confidence

Nemotron 3 Super 120B-A12B

Open weights · Medium · Published size · 120B total · 12B active

NVIDIA publishes 120B total/12B active. Public BF16 checkpoint exists under the full NVIDIA-prefixed repository name; the leaderboard’s shorter HF URL failed retrieval. NVIDIA open model license.

• NVIDIA model card
• nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16 model card and released files

Checked 2026-10-10 · high confidence

DeepSeek-V3.1

Open weights · Large · Published size · 671B total · 37B active

Official model card reports 671B total and 37B active; released MIT checkpoint includes additional storage tensors, accounting for the larger HF inventory.

• deepseek-ai/DeepSeek-V3.1 model card and released files
• deepseek-ai/DeepSeek-V3.1 checkpoint inventory

Checked 2026-10-10 · high confidence

DeepSeek-R1-0528

Open weights · Large · Published size · 671B total · 37B active

May 28 update of R1 retains the published 671B/37B architecture; its MIT weights are present. The HF tensor count includes auxiliary tensors and is larger than the architectural count.

• deepseek-ai/DeepSeek-R1-0528 model card and released files
• deepseek-ai/DeepSeek-R1 model card and released files

Checked 2026-10-10 · high confidence

DeepSeek-R1

Open weights · Large · Published size · 671B total · 37B active

Original R1, distinct from the May 2025 update; official model table reports 671B total/37B active. Public MIT checkpoint. BrowseMaster names this version as executor.

• deepseek-ai/DeepSeek-R1 model card and released files
• deepseek-ai/DeepSeek-R1 checkpoint inventory

Checked 2026-10-10 · high confidence

OpenSeeker v1 30B SFT

Open weights · Small · Published size · 30B total · 3B active

Evaluated v1 checkpoint fine-tunes Qwen3-30B-A3B-Thinking-2507. Public MIT weights verified. Synthesis teacher is training-only; runtime Visit summarizer identity is tracked separately.

• PolarSeeker/OpenSeeker-v1-30B-SFT model card and released files
• OpenSeeker paper

Checked 2026-10-10 · high confidence

OpenSeeker undisclosed Visit summarizer

Uncertain weights · Medium · Estimated

A webpage summary model is required, but evaluated identity and weights are not published in the inspected source/configuration. Medium is a low-confidence provisional category informed by the related WebSailor pipeline’s 72B summary helper; it is not an asserted parameter count or confirmed model identity. This is the runtime Visit helper; the training-only historical-context teacher is excluded.

• OpenSeeker runtime tool calls
• OpenSeeker Visit implementation
• WebSailor inference instructions

Checked 2026-10-10 · low confidence

o4-mini

Closed weights · Large · Price proxy

Token price: $1.1 input / $4.4 output per million. Blended 3:1: $1.925/M.

Parameter count is undisclosed. Large is a price proxy from standard uncached API text rates; it does not assert a physical parameter size. Tool fees and cached, batch, flex, and priority rates are excluded.

• OpenAI o4-mini API model documentation

Checked 2026-10-10 · high confidence

GLM-4.5

Open weights · Large · Published size · 355B total · 32B active

Published 355B total/32B active architecture, MIT weights. Stored tensor count is approximately 358.3B; the category is unchanged.

• zai-org/GLM-4.5 model card and released files
• zai-org/GLM-4.5 checkpoint inventory

Checked 2026-10-10 · high confidence

GLM-4.5-Air

Open weights · Medium · Published size · 106B total · 12B active

Published 106B total/12B active architecture, MIT weights. Stored tensor count is approximately 110.5B; the category is unchanged.

• zai-org/GLM-4.5-Air model card and released files
• zai-org/GLM-4.5-Air checkpoint inventory

Checked 2026-10-10 · high confidence

Perplexity Agent Pro historical stack

Uncertain weights · Large · Estimated

Historical June 2026 Pro evaluation did not pin its underlying model. Large is a provisional inference from the multi-step flagship-model preset at launch. Current dynamic presets have changed and cannot verify the historical stack; per-query costs are not token prices.

• Exa Agent launch
• Perplexity preset configuration and versioning
• Agent API launch

Checked 2026-10-10 · low confidence

WebExplorer-8B

Open weights · Small · Published size · 8B total

Qwen3-8B-derived RL checkpoint is released under Apache-2.0. The 15.7% result uses Avg@4 and an LLM judge; scoring judges are not runtime backbone models.

• hkust-nlp/WebExplorer-8B model card and released files
• WebExplorer paper

Checked 2026-10-10 · high confidence

InfoAgent-14B

Uncertain weights · Small · Published size · 14B total

Evaluated Qwen3-14B post-trained model. The current Microsoft HF repository contains RE-TRAC-30B-A3B weights, not this exact 14B checkpoint. Do not treat the later release as proof of the evaluated checkpoint’s availability.

• InfoAgent paper, especially appendix B.2
• microsoft/InfoAgent repository
• microsoft/InfoAgent model card and released files

Checked 2026-10-10 · medium confidence

GPT-4o mini

Closed weights · Medium · Price proxy

Token price: $0.15 input / $0.6 output per million. Blended 3:1: $0.2625/M.

Standard uncached input/output pricing gives a $0.2625/M blend: Medium price proxy. InfoAgent uses this hosted model for search snippets. Parameter count is undisclosed.

• GPT-4o mini model documentation

Checked 2026-10-10 · high confidence

Qwen3 Embedding 0.6B

Open weights · Tiny · Published size · 0.6B total

Public Apache-2.0 embedding checkpoint, used in InfoAgent’s retrieval environment.

• Qwen/Qwen3-Embedding-0.6B model card and released files

Checked 2026-10-10 · high confidence

Qwen3 Reranker 0.6B

Open weights · Tiny · Published size · 0.6B total

Public Apache-2.0 reranker checkpoint, used in InfoAgent’s retrieval environment.

• Qwen/Qwen3-Reranker-0.6B model card and released files

Checked 2026-10-10 · high confidence

DeepDive-32B

Uncertain weights · Small · Published size · 32B total

The authors’ repository identifies a 32B trained model but its model table still marks Hugging Face release as coming soon. Exact checkpoint availability remains unverified. OpenReview PDF retrieval encountered browser verification.

• THUDM/DeepDive repository
• DeepDive paper (retrieval encountered browser verification)

Checked 2026-10-10 · medium confidence

Exa Research Pro historical stack

Uncertain weights · Large · Estimated

The 2025 Research Pro row is a hosted research system, without a pinned backbone or token price. Large is provisional based on Exa’s documented frontier research-model approach, which is not proof of this earlier stack. The historical score and product identity are confirmed by Parallel’s dated September 2025 report.

• Parallel September 2025 benchmark report
• Exa Agent launch

Checked 2026-10-10 · low confidence

WebSailor-72B

Uncertain weights · Medium · Published size · 72B total

Paper identifies 72B evaluated backbone. No official 72B checkpoint was found; the parent HF repository has no weight shards and says coming soon. Released 7B and 32B versions do not prove 72B availability.

• WebSailor paper, appendix A.1
• Alibaba-NLP/WebSailor model card and released files

Checked 2026-10-10 · medium confidence

Qwen2.5-72B-Instruct

Open weights · Medium · Published size · 72.7B total

Published total is 72.7B, with 70B non-embedding parameters. Public weights under the Qwen license. WebSailor’s evaluation instructions explicitly name this as summary model.

• Qwen/Qwen2.5-72B-Instruct model card and released files
• WebSailor inference instructions

Checked 2026-10-10 · high confidence

WebSailor-32B

Open weights · Small · Published size · 32B total

Public Apache-2.0 checkpoint verified. The evaluation also uses Qwen2.5-72B-Instruct for webpage summaries; classify the complete runtime stack separately.

• Alibaba-NLP/WebSailor-32B model card and released files
• WebSailor paper, appendix A.1

Checked 2026-10-10 · high confidence

Qwen2.5-72B-Instruct

Open weights · Medium · Published size · 72.7B total

Published total is 72.7B, with 70B non-embedding parameters. Public weights under the Qwen license. WebSailor’s evaluation instructions explicitly name this as summary model.

• Qwen/Qwen2.5-72B-Instruct model card and released files
• WebSailor inference instructions

Checked 2026-10-10 · high confidence

Perplexity Agent Fast historical stack

Uncertain weights · Medium · Estimated

June 2026 Fast preset is not pinned. Medium is a provisional estimate based on the provider’s low-cost, single-lookup preset role, not on the $0.005/query figure. Current preset models are mutable and do not prove historical identity or weights availability.

• Exa Agent launch
• Perplexity preset configuration and versioning
• Agent API launch

Checked 2026-10-10 · low confidence

OpenAI o1

Closed weights · Large · Price proxy

Token price: $15 input / $60 output per million. Blended 3:1: $26.25/M.

Published standard uncached rates for the original full o1 model; no parameter count published. Original BrowseComp baseline has no browsing tools.

• OpenAI o1 model documentation

Checked 2026-10-10 · high confidence

DeepSeek-R1-0528

Open weights · Large · Published size · 671B total · 37B active

May 28 update of R1 retains the published 671B/37B architecture; its MIT weights are present. The HF tensor count includes auxiliary tensors and is larger than the architectural count.

• deepseek-ai/DeepSeek-R1-0528 model card and released files
• deepseek-ai/DeepSeek-R1 model card and released files

Checked 2026-10-10 · high confidence

Claude Opus 4.1

Closed weights · Large · Price proxy

Token price: $15 input / $75 output per million. Blended 3:1: $30/M.

Standard input/output prices, excluding cache and batch discounts. Exact parameter count and model weights are not public. Historical Parallel test is distinct from Anthropic’s own evaluation.

• Claude standard token pricing

Checked 2026-10-10 · high confidence

WebSailor-7B

Open weights · Small · Published size · 7B total

Public Apache-2.0 checkpoint verified. The evaluation also uses Qwen2.5-72B-Instruct for webpage summaries; classify the complete runtime stack separately.

• Alibaba-NLP/WebSailor-7B model card and released files
• WebSailor paper, appendix A.1

Checked 2026-10-10 · high confidence

Qwen2.5-72B-Instruct

Open weights · Medium · Published size · 72.7B total

Published total is 72.7B, with 70B non-embedding parameters. Public weights under the Qwen license. WebSailor’s evaluation instructions explicitly name this as summary model.

• Qwen/Qwen2.5-72B-Instruct model card and released files
• WebSailor inference instructions

Checked 2026-10-10 · high confidence

Perplexity Sonar Deep Research historical stack

Uncertain weights · Large · Estimated

Legacy 2025 Sonar research stack has no published parameter count in retrieved sources. Large is provisional based on its long-reasoning research configuration. Historical token rates could not be reverified: old model/pricing pages now redirect to the newer Agent API. No replacement model is substituted.

• Sonar migration documentation
• Perplexity historical changelog
• Parallel September 2025 benchmark report

Checked 2026-10-10 · low confidence

GPT-4o Search Preview (2025-03-11)

Closed weights · Large · Price proxy

Token price: $2.5 input / $10 output per million. Blended 3:1: $4.375/M.

The BrowseComp paper section 4.1 pins GPT-4o with browsing to this specialized search model. Official documentation lists the exact snapshot, now deprecated, with standard uncached text prices of $2.50 input/$10 output per million tokens. The $4.375/M blend is a Large price proxy. Search tool-call fees are excluded; parameter count is undisclosed.

• GPT-4o Search Preview model documentation and snapshot pricing
• BrowseComp paper, section 4.1: evaluated model snapshots

Checked 2026-10-10 · high confidence

GPT-4.5

Closed weights · Large · Price proxy

Token price: $75 input / $150 output per million. Blended 3:1: $93.75/M.

Published standard uncached rates for GPT-4.5 Preview (2025-02-27); API is now deprecated. Parameter count and weights are not published.

• GPT-4.5 Preview model documentation

Checked 2026-10-10 · high confidence

GPT-4o (2024-08-06)

Closed weights · Large · Price proxy

Token price: $2.5 input / $10 output per million. Blended 3:1: $4.375/M.

The BrowseComp paper section 4.1 pins the no-browsing baseline to gpt-4o-2024-08-06. Official GPT-4o documentation lists this snapshot with standard uncached text pricing of $2.50 input/$10 output per million tokens. The $4.375/M blend is a Large price proxy; parameter count is undisclosed.

• GPT-4o model documentation
• BrowseComp paper, section 4.1: evaluated model snapshots

Checked 2026-10-10 · high confidence