Benchmark profile

Software Engineering Benchmark Verified (SWE-bench Verified)

A curated, human-verified subset of SWE-bench that tests models on resolving real GitHub issues from popular open-source Python repositories like Django, Flask, and scikit-learn.

Data verified July 20, 2026

Claude Mythos 5 leads the SWE-bench Verified leaderboard on BenchLM's July 2026 update with 95.5%, ahead of Claude Fable 5 (95%) and Claude Opus 4.8 (88.6%), across 58 tracked models.

Top models on SWE-bench Verified — July 20, 2026

As of July 20, 2026, Claude Mythos 5 leads the SWE-bench Verified leaderboard with 95.5% , followed by Claude Fable 5 (95%) and Claude Opus 4.8 (88.6%).

1Closed

Claude Mythos 5

Anthropic

claude-mythos-5

95.5%

Overall 83.93Context 1M+

2Closed

Claude Fable 5

Anthropic

claude-fable-5

95%

Overall 83.68Context 1M+

3Closed

Claude Opus 4.8

Anthropic

claude-opus-4-8

88.6%

Overall 78.34Context 1M

58 modelsCoding16% of category scoreRefreshingUpdated July 20, 2026

Leaderboard (58 models)

Score

Claude Mythos 5Anthropic · Closed

95.5%

Claude Fable 5Anthropic · Closed

95%

Claude Opus 4.8Anthropic · Closed

88.6%

Claude Opus 4.7 (Adaptive)Anthropic · Closed

87.6%

Claude Sonnet 5Anthropic · Closed

85.2%

GPT-5.3 CodexOpenAI · Closed

85%

Ornith-1.0-397BDeepReinforce AI · Open weight

82.4%

Claude Opus 4.5Anthropic · Closed

80.9%

Claude Opus 4.6Anthropic · Closed

80.8%

DeepSeek V4 Pro (Max)DeepSeek · Open weight

80.6%

MiniMax M3MiniMax · Open weight

80.5%

Qwen3.7 MaxAlibaba · Closed

80.4%

Kimi K2.6Moonshot AI · Open weight

80.2%

GPT-5.2OpenAI · Closed

80%

Claude Sonnet 4.6Anthropic · Closed

79.6%

DeepSeek V4 Pro (High)DeepSeek · Open weight

79.4%

DeepSeek V4 Flash (Max)DeepSeek · Open weight

79%

Qwen3.6 PlusAlibaba · Closed

78.8%

DeepSeek V4 Flash (High)DeepSeek · Open weight

78.6%

MiMo-V2-ProXiaomi · Closed

78%

GLM-5Z.AI · Open weight

77.8%

Qwen3.7 PlusAlibaba · Closed

77.7%

InklingThinking Machines Lab · Open weight

77.6%

Mistral Medium 3.5 128BMistral · Open weight

77.6%

Muse SparkMeta · Closed

77.4%

Qwen3.6-27BAlibaba · Open weight

77.2%

Claude Sonnet 4.5Anthropic · Closed

77.2%

Kimi K2.5Moonshot AI · Open weight

76.8%

Kimi K2.5 (Reasoning)Moonshot AI · Closed

76.8%

Grok 4.20xAI · Closed

76.7%

Qwen3.5 397BAlibaba · Open weight

76.2%

Ornith-1.0-35BDeepReinforce AI · Open weight

75.6%

MiMo-V2-OmniXiaomi · Closed

74.8%

Laguna M.1Poolside · Closed

74.6%

Claude 4.1 OpusAnthropic · Closed

74.5%

Hy3 PreviewTencent · Open weight

74.4%

GLM-4.7Z.AI · Open weight

73.8%

DeepSeek V4 FlashDeepSeek · Open weight

73.7%

DeepSeek V4 ProDeepSeek · Open weight

73.6%

MAI-Thinking-1Microsoft · Closed

73.5%

Qwen3.6-35B-A3BAlibaba · Open weight

73.4%

MiMo-V2-FlashXiaomi · Open weight

73.4%

Claude Haiku 4.5Anthropic · Closed

73.3%

Claude 4 SonnetAnthropic · Closed

72.7%

Qwen3.5-27BAlibaba · Open weight

72.4%

Qwen3.5-122B-A10BAlibaba · Open weight

72%

Nemotron 3 UltraNVIDIA · Open weight

71.9%

Grok Code Fast 1xAI · Closed

70.8%

Laguna XS.2Poolside · Open weight

69.9%

Ornith-1.0-9BDeepReinforce AI · Open weight

69.4%

Qwen3.5-35B-A3BAlibaba · Open weight

69.2%

Gemini 2.5 ProGoogle · Closed

63.8%

GPT-4.1OpenAI · Closed

54.6%

ZAYA1-74B-PreviewZyphra · Open weight

53.2%

o3-miniOpenAI · Closed

49.3%

Claude 3.5 SonnetAnthropic · Closed

49%

DeepSeek V3DeepSeek · Open weight

42%

GPT-4.1 miniOpenAI · Closed

23.6%

According to BenchLM.ai, Claude Mythos 5 leads the SWE-bench Verified benchmark with a score of 95.5%, followed by Claude Fable 5 (95%) and Claude Opus 4.8 (88.6%). The scores show moderate spread, with meaningful differences between the top tier and mid-tier models.

58 models have been evaluated on SWE-bench Verified. The benchmark falls in the Coding category. This category carries a 20% weight in BenchLM.ai's overall scoring system. Within that category, SWE-bench Verified contributes 16% of the category score, so strong performance here directly affects a model's overall ranking.

About SWE-bench Verified

Year

2024

Tasks

500 verified issues

Format

Code patch generation

Difficulty

Professional software engineering

SWE-bench Verified is the gold standard for evaluating AI coding agents on real-world software engineering tasks. Each task requires understanding codebases, writing patches, and passing test suites.

SWE-bench: Can Language Models Resolve Real-World GitHub Issues?

BenchLM freshness & provenance

Version

SWE-bench Verified 2024

Refresh cadence

Annual

Staleness state

Refreshing

Question availability

Public benchmark set

Refreshing

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does SWE-bench Verified measure?

A curated, human-verified subset of SWE-bench that tests models on resolving real GitHub issues from popular open-source Python repositories like Django, Flask, and scikit-learn.

Which model scores highest on SWE-bench Verified?

Claude Mythos 5 by Anthropic currently leads with a score of 95.5% on SWE-bench Verified.

How many models are evaluated on SWE-bench Verified?

58 AI models have been evaluated on SWE-bench Verified on BenchLM.

Compare Top Models on SWE-bench Verified

Claude Mythos 5 vs Claude Fable 5 Claude Fable 5 vs Claude Opus 4.8 Claude Opus 4.8 vs Claude Opus 4.7 (Adaptive)Claude Opus 4.7 (Adaptive) vs Claude Sonnet 5

Learn More

Read our explainer: SWE-bench Verified benchmark deep dive

Last updated: July 20, 2026 · BenchLM version SWE-bench Verified 2024

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.