Skip to content

Benchmark

SWE-bench Verified

Coding benchmarks compare models on programming tasks. Scores are percentages. Higher is better.

Models measured
130
Scores recorded
205
Sources
5
IQ weight
2.70

Score distribution

70%
Mean
73.5%
Median
13.7%
Std dev
51.6%
10th percentile
80.8%
90th percentile
8.7%
Min
8.7%96%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.70 for chat models, measured over 129 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 5· Steel.dev Agent Leaderboards data96%
2Claude Fable 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
95%
3Claude Fable 5Fable/Mythos 5 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.95%
4Ember-1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim92.2%
5Claude Opus 4.8· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
88.6%
6Claude Opus 4.8Fable/Mythos 5 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.88.6%
7Claude Opus 4.8Opus 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.88.6%
8Claude Opus 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
87.6%
9Claude Opus 4.7adaptive thinking · max effort · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.87.6%
10Claude 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.87.6%
11Claude Sonnet 5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.2%
12Claude Sonnet 5adaptive thinking · max effort · 5 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85.2%
13GPT-5.6 Sol· Steel.dev Agent Leaderboards data82.2%
14Inkling 2 Small· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82.2%
15Claude Sonnet 4.5parallel test-time compute· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.82%
16Beam 1.0 501B A23B· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
81.6%
17Claude Opus 4.6Anthropic Opus 4.6 launch page · adaptive thinking · max effort unless noted · prompt modification · 25 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.81.4%
18Claude Opus 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80.9%
19Claude Opus 4.5no extended thinking · 200K context · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.9%
20Claude Opus 4.5Opus 4.6 System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.9%
21Claude Opus 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80.8%
22Claude Opus 4.6Opus 4.6 System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.8%
23Claude Opus 4.564K thinking budget · 200K context · avg@5· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.6%
24DeepSeek V4 Pro· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim80.6%
25Gemini 3.1 Pro Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80.6%
26Gemini 3.1 Pro PreviewFable/Mythos 5 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.80.6%
27Gemini 3.1 Pro PreviewOpus 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.80.6%
28Gemini 3.1 Pro Previewsingle attempt· Reported by Google DeepMind· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.6%
29MiniMax M3· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80.5%
30MiniMax M3Claude Code · four-run mean· Reported by MiniMax Research· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.5%
31Qwen3.7 Max· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim80.4%
32Qwen3.7 MaxQwen3.7-Max release · internal bash + file-edit agent scaffold · temp 1.0 · top_p 0.95 · 200K· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.4%
33Claude Sonnet 4parallel test-time compute · regression-test rejection · internal scoring model selects one patch· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
34Claude Sonnet 4.6prompt-modified ablation · >100 tool calls · own tests first· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
35Inkling Small· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80.2%
36Kimi K2.6· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 3 sources
80.2%
37Kimi K2.6Kimi K2.6 official benchmark table · thinking · temp 1.0 · top_p 1.0 · 262,144 context · in-house SWE-agent-derived harness · avg@10· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
38MiniMax M2.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 4 sources
80.2%
39MiniMax M2.5main capability figure· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
40GLM 5.2· Steel.dev Agent Leaderboards data80%
41GPT 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
80%
42GPT 5.2Opus 4.6 System Card capability summary · adaptive thinking · max effort · default sampling · 5 trials unless noted· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80%
43GPT 5.2xhigh· Reported by OpenAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80%
44MiniMax M2.5official prose · Droid · default prompt· Reported by MiniMax· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.79.7%
45Claude Sonnet 4.6· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
79.6%
46Claude Sonnet 4.6adaptive thinking · max effort · 10 trials· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.79.6%
47Claude Opus 4parallel test-time compute · regression-test rejection · internal scoring model selects one patch· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.79.4%
48Inkling 2 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim79.2%
49DeepSeek V4 Flash· Hugging Face Official Benchmarks data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
79%
50DeepSeek 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.79%

Showing 1–50 of 205 results.

Data credits: Data by LLM Stats · Hugging Face official benchmark leaderboards; licence varies by benchmark · 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. · Steel.dev Agent Leaderboards, MIT (licence) · OpenEvals/leaderboard-data, MIT (licence)