Benchmark
BrowseComp
Agentic benchmarks measure how well models carry out tasks that take several steps. Scores are percentages. Higher is better.
- Models measured
- 84
- Scores recorded
- 203
- Sources
- 4
- IQ weight
- 2.43
Score distribution
- 66.4%
- Mean
- 74.7%
- Median
- 21.8%
- Std dev
- 35.2%
- 10th percentile
- 90.1%
- 90th percentile
- 0.6%
- Min
This benchmark has a weight of 2.43 for chat models, measured over 84 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.
Leaderboard
| # | Model | Score |
|---|---|---|
| 51 | GPT 5.5Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.4% |
| 52 | GPT 5.5xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.4% |
| 53 | Claude Opus 4.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 84.3% |
| 54 | Claude Opus 4.8OpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.3% |
| 55 | Claude Opus 4.8Opus 5 System Card capability summary · adaptive thinking · max effort unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.3% |
| 56 | Claude Opus 4.8single-agent· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.3% |
| 57 | Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 84.2% |
| 58 | Hy3 (Deep Infra)Hy3 official Benchmark Appendix · highest reasoning tier unless noted· Reported by Tencent Hunyuan· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84.2% |
| 59 | Claude Opus 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 84% |
| 60 | Claude Opus 4.6Anthropic Opus 4.6 launch page · adaptive thinking · max effort unless noted · single-agent· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 84% |
| 61 | MiniMax M3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.5% |
| 62 | MiniMax M3WebExplorer-style agent · history discarded beyond 64K· Reported by MiniMax Research· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.5% |
| 63 | MiniMax M3official model-page overview figure · rounded· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.5% |
| 64 | DeepSeek V4 Pro 0813Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · in-house search/visit/python harness · 500 steps · discard-all context strategy· Reported by Shanghai AI Laboratory· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.4% |
| 65 | GPT-5.6 Luna· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 83.3% |
| 66 | GPT-5.6 LunaOpenAI GPT-5.6 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.3% |
| 67 | Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context · context management · discard-all strategy· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 83.2% |
| 68 | GPT 5.4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 82.7% |
| 69 | GPT 5.4OpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 82.7% |
| 70 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, perplexity, server-tool, 25, high· OpenRouter benchmark exports | 82.4% |
| 71 | Claude Opus 5OpenRouter harness, anthropic/claude-opus-5, exa, server-tool, 25, high· OpenRouter benchmark exports | 82.2% |
| 72 | Claude Sonnet 4.6multi-agent orchestrator · subagents with tools · compaction · max effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 82.1% |
| 73 | Ling 3.0 Flash SanteLing-3.0-flash-Sante official evaluation image · highest reasoning tier · temperature 1.0 / top_p 0.95 unless otherwise stated · exact API snapshot and run count not published · single agent with context · 256K · up to 1,000 turns · discard-all at 40% context usage· Reported by Ant inclusionAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 80.2% |
| 74 | Claude Opus 4.7Opus 4.8 System Card capability summary · adaptive thinking · max effort · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 79.8% |
| 75 | Claude Opus 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 79.3% |
| 76 | Claude Opus 4.7web search/fetch + code · thinking off · max effort · 10M total tokens · compaction at 200K· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 79.3% |
| 77 | GLM 5.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 79.3% |
| 78 | GLM 5.1GLM-5.1 official release table · with context management· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 79.3% |
| 79 | Inkling 2 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 79% |
| 80 | Qwen3.5 397B A17B· Steel.dev Agent Leaderboards dataValues from 2 sources
| 78.6% |
| 81 | Qwen3.5 397B A17BQwen3.5 release/model card · discard-all strategy used by DeepSeek-V3.2 and Kimi K2.5· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 78.6% |
| 82 | Kimi K2.5· Steel.dev Agent Leaderboards dataValues from 2 sources
| 78.4% |
| 83 | Kimi K2.5Agent Swarm· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 78.4% |
| 84 | NVIDIA Nemotron 3.5 Super VL 120B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 78% |
| 85 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, parallel, server-tool, 25, high· OpenRouter benchmark exports | 77.9% |
| 86 | GPT-5.2 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 77.9% |
| 87 | GPT-5.2 ProOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 77.9% |
| 88 | GPT-5.2 Prostandard tool-use setup · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 77.9% |
| 89 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, exa, server-tool, 25, high· OpenRouter benchmark exports | 77.8% |
| 90 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, native, server-tool, 25, high· OpenRouter benchmark exports | 77.8% |
| 91 | Inkling Small· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 77.4% |
| 92 | Doubao Seed 2.0 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 77.3% |
| 93 | GPT-5.3 Codex· Steel.dev Agent Leaderboards data | 77.3% |
| 94 | GPT-5.3 CodexOpenAI GPT-5.4 launch table· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 77.3% |
| 95 | Inkling FP4· Steel.dev Agent Leaderboards data | 77.1% |
| 96 | DeepSeek V4 Flash 0731OpenRouter harness, deepseek/deepseek-v4-flash-0731, perplexity, server-tool, 25, high· OpenRouter benchmark exports | 77% |
| 97 | Beam 1.0 501B A23B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 76.8% |
| 98 | MiniMax M2.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 76.3% |
| 99 | MiniMax M2.5main capability figure · with context· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 76.3% |
| 100 | Claude Sonnet 4.6· Steel.dev Agent Leaderboards dataValues from 2 sources
| 76.2% |
Showing 51–100 of 203 results.
Data credits: Steel.dev Agent Leaderboards, MIT (licence) · Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence)