BenchmarkActivecategory · codingfamily · swe-bench · variant Verified

SWE-bench Verified

swebench.com

resolve real GitHub issues (500 human-validated instances)

data quality57

Updated 5 h ago · first seen 11 Sept 2026

Metric: resolved · % ↑
Current results: 158
Models: 82
Current leader: Claude 4 Sonnet 76.8%

Filters

Comparability group

Trust level

Views

Frontier over time · resolved · board=Verified · system=EPAM AI/Run Developer Agent

76.8%Claude 4 Sonnet Anthropic Community4 Aug 2025
62.8%claude-35-sonnet Anthropic Community28 Feb 2025
55.4%claude-35-sonnet Anthropic Community12 Dec 2024
39.6%claude-35-sonnet Anthropic Community29 Oct 2024
27%gpt-4o OpenAI Community16 Oct 2024
24%gpt-4o OpenAI Community20 Aug 2024

Includes closed rows (history). A point is emitted whenever a result beats every earlier result of the same group, ordered by evaluated_at when the source publishes it, else observed_at.

Leaderboard 3 models

Select models with +, then Compare.

Leaderboard
#	Model	Score	Trust	Configuration	vs leader	Evaluated	Source	Actions
1	Claude 4 SonnetAnthropic · Claude 4	76.8%	Community	group defaults	leader	4 Aug 2025	swebench.comT2	History
2▼1	claude-35-sonnetClosedAnthropic · Claude 35	62.8%	Community	group defaults	Partially comparable-14.0 pt	28 Feb 2025	swebench.comT2	History
3▼1	gpt-4oClosedOpenAI · GPT 4	27%	Community	group defaults	Partially comparable-49.8 pt	16 Oct 2024	swebench.comT2	History

3 results

One row per canonical model — its best current row inside this comparability group (effort variants are folded into the model). Bars are relative to the page's best score. “vs leader” reads comparability: partially comparable = same task, conditions differ (reasoning effort, temperature, judge). Rules →

Definition

Task: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 15 h agomedium
Metric: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 15 h agomedium
Metric range: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 5 h agomedium
Variant: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 5 h agomedium
Harness: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 5 h agomedium
Comparability note: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 5 h agomedium
Known limitations: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 15 h agomedium
Paper: Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 15 h agomedium
Website: swebench.com
Source:AI Atlas curated registry (YAML, versioned in git, every entry carries its source URL)T2observed 15 h agomedium

Active group

resolved · board=Verified · system=EPAM AI/Run Developer Agent

3 models · 3 current rows

Comparability

Comparable same task and conditions · Partially comparable same task, conditions differ (effort, temperature, judge) · Not comparable different variant or metric

Also known as

SWE-bench, SWE-bench Verified

slug swe-bench-verified · bench_01M293SPEGA1XMKH0SMTFTCK2R

Rows are grouped by comparability group = canonical metric × config_key (hash of the task-defining configuration keys: variant, evaluator, harness, shots, pass regime…). A leaderboard shows one row per canonical model: its best current row inside the group. Reasoning effort, temperature or judge differences keep rows in the same group but mark them partially comparable. delta_rank compares with the ranking built from the closed (previous) rows of the same group.