Leaderboard
Agent scope| System / Submission | Score | Organization | Reported | Source |
|---|---|---|---|---|
| WebTactix (DeepSeek v3.2) Project page reports 594/812 correct tasks, or 74.34%; this is a WebTactix system using DeepSeek v3.2, not a bare model result. | 74.3% | WebTactix | Source | |
| OpAgent Open-source multi-agent browser system; repository publishes the WebArena result and final trajectories. | 71.6% | CodeFuse AI | Source | |
| ColorBrowserAgent Human-like browser agent result; code and trajectories are public in the MadeAgents repository. | 71.2% | MadeAgents | Source | |
| Claude Code + GBOX MCP Claude Code backed by GBOX MCP scaffolding; public report and trajectories are in the BabelCloud WebArena repository. | 68.0% | GBOX AI | Source | |
| DeepSky Agent Public tracker row; no separate public source or repository link was found in the tracker. | 66.9% | DeepSky | Source | |
| Narada AI Narada blog and public tracker report the WebArena result; no public code repository was found. | 64.2% | Narada AI | Source | |
| IBM CUGA CUGA reports 61.7% on WebArena and publishes code, evaluation artifacts, and results. | 61.7% | IBM | Source | |
| Kimi K2.5 Technical report reports GUI-based browsing without external tools; Kimi K2.5 weights are publicly released. | 58.9% | Moonshot AI | Source | |
| OpenAI Operator OpenAI Computer-Using Agent result reported with the Operator system card and public WebArena tracker. | 58.1% | OpenAI | Source | |
| Jace.AI (AWA-1.5) Jace reports AWA 1.5 at 57.14% on WebArena; implementation is proprietary. | 57.1% | Jace AI | Source | |
| Plan-MCTS + GPT-5-mini Plan-MCTS paper reports 55.3% on the full 812-task WebArena test set with GPT-5-mini. | 55.3% | Academic Research | Source | |
| WebOperator + GPT-4o WebOperator paper and project report 54.6% with GPT-4o; code and experiment artifacts are public. | 54.6% | KAIST KAG NLP | Source | |
| ScribeAgent + GPT-4o ScribeAgent reports 53.0% with GPT-4o; the tracker marks data as proprietary. | 53.0% | Academic Research | Source | |
| 52.1% | Academic Research | Source | ||
| Learn-by-Interact Paper reports 48.0% on WebArena; no public implementation repository was found. | 48.0% | Academic Research | Source | |
| WebUncertainty + GPT-4-Turbo New WebUncertainty reports 46.9% with GPT-4-Turbo on WebArena and publishes the project repository. | 46.9% | Academic Research | Source | |
| AgentOccam-Judge AgentOccam-Judge public tracker row; Amazon Science publishes the AgentOccam implementation. | 45.7% | Amazon Science | Source | |
| A3-Qwen3.5-9B New BrowserGym leaderboard discussion reports 42.1%; the paper/model card table reports 41.5 under its project setup. | 42.1% | McGill NLP | Source | |
| GUI-API Hybrid Agent Latest arXiv version reports 38.9%; the public tracker still lists 35.8 for an earlier or alternate setup. | 38.9% | Academic Research | Source | |
| WebPilot Paper reports 37.2%; the public tracker notes no open-source code or trajectories were released. | 37.2% | Academic Research | Source | |
| Agent Workflow Memory Workflow-memory agent result; public repository is available with the paper artifacts. | 35.5% | Academic Research | Source | |
| SteP Public tracker reports 33.5%; the SteP project publishes code and WebArena materials. | 33.5% | ASAPP Research | Source | |
| TTI Test-time interaction result; WebArena artifacts are public on Hugging Face. | 26.1% | Academic Research | Source | |
| BrowserGym + GPT-4 BrowserGym/WorkArena paper reports this WebArena baseline; BrowserGym is open source. | 23.5% | ServiceNow Research | Source | |
| AgentTrek-1.0-32B AgentTrek project reports 22.4%; code, model, and data links are public. | 22.4% | xLang AI | Source | |
| GPT-4 + Auto Eval Automatic evaluator study reports a GPT-4 WebArena result; no separate implementation repository was found. | 20.2% | Academic Research | Source | |
| GPT-4o + Tree Search Tree-search agent result from the Search Agents project page. | 19.2% | Academic Research | Source | |
| AutoWebGLM AutoWebGLM paper reports 18.2%; official repository includes WebArena evaluation materials. | 18.2% | THUDM | Source | |
| NNetNav NNetNav result with public model checkpoint linked from the tracker. | 16.3% | Stanford NLP | Source | |
| gpt-4-0613 (no not-achievable hint) Original WebArena baseline without providing the not-achievable task hint. | 14.9% | OpenAI | Source | |
| gpt-4o-2024-05-13 Public tracker row from the WebArena team with the not-achievable task hint provided. | 13.05% | OpenAI | Source | |
| gpt-4-0613 (with not-achievable hint) Original WebArena GPT-4 baseline when the not-achievable task hint is provided. | 11.7% | OpenAI | Source | |
| Patel et al. + GPT-4 Patel et al. report a GPT-4 WebArena evaluation row in the public tracker. | 9.36% | Academic Research | Source | |
| gpt-3.5-turbo-16k-0613 Original WebArena GPT-3.5 baseline. | 8.87% | OpenAI | Source | |
| Qwen-1.5-chat-72b Patel et al. tracker row for the open-weight Qwen 1.5 chat model. | 7.14% | Qwen | Source | |
| Gemini Pro WebArena tracker baseline for Gemini Pro; implementation is proprietary. | 7.12% | Source | ||
| Llama3-chat-70b Public tracker baseline for the open-weight Llama 3 70B Instruct model. | 7.02% | Meta | Source | |
| Synatra-CodeLLama7b Synatra CodeLlama-7B WebArena row from the paper and public tracker. | 6.28% | Academic Research | Source | |
| Lemur-chat-70b Lemur-chat-70B WebArena row; open model weights are available on Hugging Face. | 5.3% | OpenLemur | Source | |
| Agent-FLAN Agent-FLAN result from the paper and public tracker; code and model artifacts are public. | 4.68% | InternLM | Source | |
| CodeLlama-instruct-34b Lemur paper baseline for the public Code Llama 34B Instruct model. | 4.06% | Meta | Source | |
| AgentLM-70b AgentTuning paper reports the AgentLM-70B WebArena result and links released AgentLM models. | 3.81% | THUDM | Source | |
| Llama3-chat-8b Public tracker baseline for the open-weight Llama 3 8B Instruct model. | 3.32% | Meta | Source | |
| CodeAct Agent CodeAct paper reports a WebArena row; official CodeAct repository is public. | 2.3% | Academic Research | Source | |
| AgentLM-13b AgentTuning paper reports the AgentLM-13B WebArena result and links released AgentLM models. | 1.6% | THUDM | Source | |
| Mixtral-8x7B Public tracker baseline for the open-weight Mixtral-8x7B Instruct model. | 1.39% | Mistral AI | Source | |
| AgentLM-7b AgentTuning paper reports the AgentLM-7B WebArena result and links released AgentLM models. | 0.74% | THUDM | Source | |
| FireAct FireAct WebArena row from the public tracker; no public WebArena implementation artifact was found. | 0.25% | Academic Research | Source | |
| CodeLlama-instruct-7b Public tracker baseline for the public Code Llama 7B Instruct model. | 0.0% | Meta | Source |
About this benchmark
WebArena evaluates browser agents in reproducible, self-hosted websites instead of the open live web. Its 812 tasks cover e-commerce, forum discussion, collaborative software development, content management, maps, and reference lookup.
The benchmark is strongest when you care about repeatable web-agent experiments: every task has a controlled start state and functional success criteria rather than a changing production website.
Because many rows come from a public community tracker, a WebArena score should be read alongside the source, submitted scaffold, observation mode, and whether the result was independently reproduced.
Controlled environments improve reproducibility, but tracker rows still vary by scaffold and submission policy.
Filtered task-set or modified-grader reports are not ranked as full WebArena results unless the row notes that setup explicitly.
Example tasks
Three public tasks quoted from benchmark sources:
- "What is the top-1 best-selling product in 2022" Citation: WebArena test config
- "Tell me the full address of all international airports that are within a driving distance of 50 km to Carnegie Mellon University" Citation: WebArena test config
- "Tell me the the number of reviews that our store received by far that mention term "disappointed"" Citation: WebArena test config
Methodology
- Primary metric is end-to-end task success rate on the WebArena task set; the original GPT-4-based baseline was 14.41% versus 78.24% human performance.
- Evaluation checks functional correctness through task-specific validators and answer checks in the hosted environment.
- Scores can change with prompt scaffolding, observation mode, browser action space, and retry or step budget.
- We prefer rows tied to WebArena's public leaderboard, papers, or repositories that include enough setup detail to reproduce the run.
Links
Related benchmarks
Compare this benchmark with related pages from the hub:
Frequently asked questions
Which system is currently best on WebArena? + -
WebTactix (DeepSeek v3.2) is the system/agent setup currently leading with a tracked score of 74.3%. This ranking reflects submitted system setups (model plus tools and policy), not just a base model. Based on our latest tracked results, last updated May 27, 2026.
What should I read into a WebArena score? + -
WebArena scores are most useful for within-benchmark ranking. Read the Notes column to understand setup context, and use the methodology section before making procurement or architecture decisions.