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 |
|---|---|---|
| 101 | GLM 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 75.9% |
| 102 | GLM 5GLM-5 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. | 75.9% |
| 103 | Agents A1· Steel.dev Agent Leaderboards data | 75.5% |
| 104 | MiniMax M2.5search/tool figure · with context· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 75.1% |
| 105 | Kimi K2.5single agent · context management· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 74.9% |
| 106 | Claude Sonnet 4.6launch-page figure · web search/fetch + code · no thinking· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 74.7% |
| 107 | Claude Sonnet 4.6single agent · 10M total sampled tokens · no thinking · max effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 74% |
| 108 | GPT-5.6 LunaOpenRouter harness, openai/gpt-5.6-luna, perplexity, server-tool, 25, xhigh· OpenRouter benchmark exports | 74% |
| 109 | DeepSeek V4 FlashDeepSeek V4 official model card · Think Max· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 73.2% |
| 110 | LongCat-Flash-Thinking· Steel.dev Agent Leaderboards dataValues from 2 sources
| 73.1% |
| 111 | Ling 3.0 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 72.2% |
| 112 | Claude Sonnet 4.6single agent · 3M total sampled tokens · no thinking · max effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 69.7% |
| 113 | DeepSeek V4 Flash 0731OpenRouter harness, deepseek/deepseek-v4-flash-0731, parallel, server-tool, 25, high· OpenRouter benchmark exports | 69.1% |
| 114 | Qwen3.5 397B A17BQwen3.5 release/model card · simple context folding · 256K· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 69% |
| 115 | Step 3.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 69% |
| 116 | Step 3.5 FlashStep 3.5 Flash official headline table · vanilla · with Context Manager· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 69% |
| 117 | Step 3.5 FlashStep 3.5 technical report · Table 17 · 200-item subset · multi-agent · 721 real steps· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 68.5% |
| 118 | GPT-5.6 LunaOpenRouter harness, openai/gpt-5.6-luna, exa, server-tool, 25, xhigh· OpenRouter benchmark exports | 68.4% |
| 119 | GLM 5.1GLM-5.1 official release table · without context management· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 68% |
| 120 | DeepSeek V3.2Thinking mode · temperature 1.0 · 128K context · Discard-all context management· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 67.6% |
| 121 | Doubao Seed 1.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim | 67.6% |
| 122 | GLM 4.7· Steel.dev Agent Leaderboards dataValues from 2 sources
| 67.5% |
| 123 | DeepSeek V4 Flash 0731OpenRouter harness, deepseek/deepseek-v4-flash-0731, exa, server-tool, 25, high· OpenRouter benchmark exports | 67.4% |
| 124 | Claude Opus 5OpenRouter harness, anthropic/claude-opus-5, perplexity, server-tool, 5, high· OpenRouter benchmark exports | 66.5% |
| 125 | Step 3.5 FlashStep 3.5 technical report · Table 17 · 200-item subset · discard-all · 302 real steps· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 66% |
| 126 | GPT 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 65.8% |
| 127 | GPT 5.2OpenAI 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. | 65.8% |
| 128 | GPT 5.2standard tool-use setup · xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 65.8% |
| 129 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, perplexity, server-tool, 5, high· OpenRouter benchmark exports | 65.2% |
| 130 | Claude Sonnet 4.6single agent · 1M total sampled tokens · no thinking · max effort· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 64.7% |
| 131 | Qwen3.5 122B-A10B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 63.8% |
| 132 | GLM 5GLM-5 official release table · without context management· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 62% |
| 133 | MiniMax M2.1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 62% |
| 134 | Qwen3.5 27B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 61% |
| 135 | Qwen3.5 35B A3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 61% |
| 136 | Kimi K2.5single agent · no context management· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60.6% |
| 137 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, parallel, server-tool, 5, high· OpenRouter benchmark exports | 60.2% |
| 138 | DeepSeek V3.2Thinking mode · temperature 1.0 · 128K context · Summary context management · average steps 140→364· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 60.2% |
| 139 | Kimi K2 Thinking· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 60.2% |
| 140 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, exa, server-tool, 5, high· OpenRouter benchmark exports | 60.1% |
| 141 | Gemini 3 Pro Preview· Steel.dev Agent Leaderboards data | 59.2% |
| 142 | GPT-5.6 SolOpenRouter harness, openai/gpt-5.6-sol, native, server-tool, 5, high· OpenRouter benchmark exports | 59% |
| 143 | MiMo V2 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 58.3% |
| 144 | MiMo V2 FlashMiMo-V2-Flash final evaluation · with context management· Reported by Xiaomi MiMo· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58.3% |
| 145 | Step 3.5 FlashStep 3.5 technical report · Table 17 · 200-item subset · keep-first-and-last-K · 244 real steps· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 58% |
| 146 | Claude Opus 5OpenRouter harness, anthropic/claude-opus-5, parallel, server-tool, 5, high· OpenRouter benchmark exports | 57.8% |
| 147 | GPT-5.6 LunaOpenRouter harness, openai/gpt-5.6-luna, perplexity, server-tool, 5, xhigh· OpenRouter benchmark exports | 57% |
| 148 | Step 3.5 FlashStep 3.5 technical report · Table 17 · 200-item subset · summary · 131 real steps· Reported by StepFun· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it. | 57% |
| 149 | GPT-5.6 LunaOpenRouter harness, openai/gpt-5.6-luna, parallel, server-tool, 25, xhigh· OpenRouter benchmark exports | 56% |
| 150 | GPT 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claimValues from 2 sources
| 54.9% |
Showing 101–150 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)