Skip to main content
Radar

Five or fewer confirmed AI changes, with original sources, on mornings when something changed.A free source-linked morning brief.

Start the free Radar Brief

DeepSWE

A long-horizon software engineering benchmark from Datacurve for measuring frontier coding agents on original tasks drawn from active open-source repositories.

Data verified 26 confirmed releases in the last 30 daysStart the free Radar Brief

How we show DeepSWE

We mirror the public DeepSWE leaderboard JSON from Datacurve. The snapshot shows the best available mini-swe-agent configuration per model, while preserving 63 underlying effort-level rows in the source metadata.

DeepSWE evaluates coding agents on 113 original, long-horizon software engineering tasks across 91 repositories and 5 languages, using isolated task environments and program-based verifiers.

We keep DeepSWE display only. Each public row combines a model, mini-swe-agent harness, and reasoning-effort setting, so the scores do not enter the weighted model-only rankings.

Snapshot

26 best-per-model rows113 tasks91 repositories5 languages63 effort rows in sourceDisplay only

Pass@1 on DeepSWE — August 26, 2026

We mirror the published pass@1 view for DeepSWE. Claude Opus 5 leads the public snapshot at 73.6%, followed by GPT-5.6 Sol (72.7%) and Claude Fable 5 (69.7%). We do not use these results to rank models overall.

26 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated August 26, 2026

Pass@1 table (26 models)

Score
1
Claude Opus 5Anthropic · Closedmini-swe-agent · max reasoning
73.6%
2
GPT-5.6 SolOpenAI · Closedmini-swe-agent · max reasoning
72.7%
3
Claude Fable 5Anthropic · Closedmini-swe-agent · max reasoning
69.7%
4
GPT-5.6 TerraOpenAI · Closedmini-swe-agent · max reasoning
69.6%
5
GLM-5.3Z.AI · Open weightmini-swe-agent · max reasoning
69.0%
6
Kimi K3Moonshot AI · Closedmini-swe-agent · max reasoning
68.5%
7
GPT-5.6 LunaOpenAI · Closedmini-swe-agent · max reasoning
67.2%
8
GPT-5.5OpenAI · Closedmini-swe-agent · xhigh reasoning
67.0%
9
Grok 4.6xAI · Closedmini-swe-agent · xhigh reasoning
66.7%
10
Gemini 3.7 FlashGoogle · Closedmini-swe-agent · high reasoning
65.3%
11
GLM-5.3-FlashZ.AI · Open weightmini-swe-agent · max reasoning
63.4%
12
DeepSeek V4 Pro 0813DeepSeek · Closedmini-swe-agent · max reasoning
62.8%
13
Claude Opus 4.8Anthropic · Closedmini-swe-agent · max reasoning
59.0%
14
Qwen3.8 MaxAlibaba · Open weightmini-swe-agent · xhigh reasoning
57.5%
15
Muse Spark 1.2Meta · Closedmini-swe-agent · xhigh reasoning
54.9%
16
Claude Sonnet 5Anthropic · Closedmini-swe-agent · max reasoning
53.8%
17
Grok 4.5xAI · Closedmini-swe-agent · high reasoning
53.8%
18
DeepSeek V4 Flash 0731DeepSeek · Closedmini-swe-agent · max reasoning
53.3%
19
Muse Spark 1.1Meta · Closedmini-swe-agent · xhigh reasoning
53.3%
20
GPT-5.4OpenAI · Closedmini-swe-agent · xhigh reasoning
51.8%
21
Gemini 3.6 FlashGoogle · Closedmini-swe-agent · high reasoning
46.7%
22
GLM-5.2Z.AI · Open weightmini-swe-agent · max reasoning
43.8%
23
Gemini 3.5 FlashGoogle · Closedmini-swe-agent · high reasoning
36.1%
24
Kimi K2.7 CodeMoonshot AI · Open weightmini-swe-agent
30.5%
25
Claude Sonnet 4.6Anthropic · Closedmini-swe-agent · high reasoning
29.9%
26
Gemini 3.1 ProGoogle · Closedmini-swe-agent · high reasoning
11.7%

The published DeepSWE snapshot places Claude Opus 5 first at 73.6%. The third row is 3.9 points behind. The broader top-10 range is 8.3 points, so many of the published results sit in a relatively narrow band.

26 models have been evaluated on DeepSWE. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. DeepSWE is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About DeepSWE

Year

2026

Tasks

113 software engineering tasks across 91 repositories and 5 languages

Format

Pass@1 with confidence interval, cost, time, and token metadata

Difficulty

Long-horizon software engineering

DeepSWE includes original tasks with isolated environments and program-based verifiers. BenchLM mirrors the public DeepSWE leaderboard JSON as display-only, using the best available mini-swe-agent configuration per model and preserving cost, time, token, and effort-level source metadata. Each row combines a model, agent harness, and reasoning-effort setting rather than a pure model-only benchmark score.

BenchLM freshness & provenance

Version

DeepSWE 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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 DeepSWE measure?

A long-horizon software engineering benchmark from Datacurve for measuring frontier coding agents on original tasks drawn from active open-source repositories.

Which model leads the published DeepSWE snapshot?

Claude Opus 5 currently leads the published DeepSWE snapshot with 73.6% pass@1. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on DeepSWE?

The August 26, 2026 contains 26 AI models.

Last updated: August 26, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 2,000+ readers.

One email each week. Unsubscribe anytime.