SWE-bench Verified Extended — Mercor evaluation (SWE-bench Verified Extended (Mercor))

We show this table for reference; we do not rank on it.

Editing repositories to resolve verified GitHub issues. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Evaluation results published by Mercor. Published evaluation aggregates only; task contents and dataset license grants are not included.

Pass@1 on SWE-bench Verified Extended (Mercor) — October 1, 2026 capture

We mirror the published pass@1 view for SWE-bench Verified Extended (Mercor). DeepSeek-V4.1-Flash leads the public snapshot at 82.70%, followed by Opus 5 (82.00%) and Fable 5 (78.70%). We do not use these results to rank models overall.

DeepSeek-V4.1-Flash

DeepSeek

Mercor extended-set run · max reasoning

Published margin ±10.3 · Original set 82.9% ±3 · 150 reported samples

82.70%

Opus 5

Anthropic

Mercor extended-set run · max reasoning

Published margin ±9.3 · Original set 93.5% ±2 · 150 reported samples

82.00%

Fable 5

Anthropic

Mercor extended-set run · max reasoning

Published margin ±9.7 · Original set 95.9% ±1.5 · 150 reported samples

78.70%

33 configurationsCodingDated Mercor configurationsDisplay onlyUpdated October 1, 2026 capture

Pass@1 table (33 configurations)

Score

DeepSeek-V4.1-FlashDeepSeekMercor extended-set run · max reasoningPublished margin ±10.3 · Original set 82.9% ±3 · 150 reported samples

82.70%

Opus 5AnthropicMercor extended-set run · max reasoningPublished margin ±9.3 · Original set 93.5% ±2 · 150 reported samples

82.00%

Fable 5AnthropicMercor extended-set run · max reasoningPublished margin ±9.7 · Original set 95.9% ±1.5 · 150 reported samples

78.70%

GPT-5.6 TerraOpenAIMercor extended-set run · max reasoningPublished margin ±11 · Original set 77.8% ±3.5 · 150 reported samples

78.70%

GPT-5.6 LunaOpenAIMercor extended-set run · max reasoningPublished margin ±11.7 · Original set 76.2% ±3.6 · 150 reported samples

74.70%

Grok 4.5xAIMercor extended-set run · high reasoningPublished margin ±10.3 · Original set 81.7% ±3.2 · 150 reported samples

70.70%

GLM-5.3ZhipuMercor extended-set run · max reasoningPublished margin ±10 · Original set 81.7% ±3.2 · 150 reported samples

68.00%

Sonnet 5AnthropicMercor extended-set run · max reasoningPublished margin ±11 · Original set 82.8% ±3 · 150 reported samples

67.30%

Opus 5.5AnthropicMercor extended-set run · max reasoningPublished margin ±11.7 · Original set 98.2% ±1.1 · 150 reported samples

66.00%

Opus 4.8AnthropicMercor extended-set run · max reasoningPublished margin ±11 · Original set 88.5% ±2.6 · 150 reported samples

63.30%

GPT-5.6 Sol (Pro)OpenAIMercor extended-set run · max reasoningPublished margin ±11.3 · Original set 72.8% ±3.4 · 150 reported samples

63.30%

Fable 5.1AnthropicMercor extended-set run · max reasoningPublished margin ±11.7 · Original set 96.6% ±1.5 · 150 reported samples

61.30%

Qwen3.8-MaxAlibabaMercor extended-set run · xhigh reasoningPublished margin ±10.7 · Original set not reported · 150 reported samples

58.70%

Opus 4.7AnthropicMercor extended-set run · max reasoningPublished margin ±11.3 · Original set 83.1% ±3 · 150 reported samples

58.00%

Fable 5.1AnthropicMercor extended-set run · high reasoningPublished margin ±12 · Original set 92.1% ±2.2 · 150 reported samples

57.30%

Sonnet 5.5AnthropicMercor extended-set run · max reasoningPublished margin ±12.7 · Original set 96.7% ±1.5 · 150 reported samples

57.30%

Gemini 3.8 FlashGoogleMercor extended-set run · high reasoningPublished margin ±12 · Original set 80.6% ±3.4 · 150 reported samples

54.00%

GLM-5.2ZhipuMercor extended-set run · max reasoningPublished margin ±12 · Original set 78.5% ±3.3 · 150 reported samples

47.30%

GPT-6.1 SolOpenAIMercor extended-set run · max reasoningPublished margin ±13.3 · Original set 87.3% ±2.8 · 150 reported samples

45.30%

GPT-5.4OpenAIMercor extended-set run · xhigh reasoningPublished margin ±12 · Original set 78.1% ±3.4 · 150 reported samples

40.00%

GPT-5.5OpenAIMercor extended-set run · xhigh reasoningPublished margin ±13 · Original set 78.8% ±3.3 · 150 reported samples

39.30%

GPT-6 LunaOpenAIMercor extended-set run · max reasoningPublished margin ±13 · Original set 76.1% ±3.5 · 150 reported samples

39.30%

GPT-6 SolOpenAIMercor extended-set run · max reasoningPublished margin ±12 · Original set 80.3% ±3.3 · 150 reported samples

39.30%

Gemini 3.6 FlashGoogleMercor extended-set run · high reasoningPublished margin ±11.7 · Original set 81.1% ±3.3 · 150 reported samples

33.30%

Gemini 3.7 FlashGoogleMercor extended-set run · high reasoningPublished margin ±10.3 · Original set 81% ±3.3 · 150 reported samples

30.00%

MiniMax-M3MiniMaxMercor extended-set run · high reasoningPublished margin ±9.7 · Original set 75.5% ±3.5 · 150 reported samples

28.70%

Sonnet 4.6AnthropicMercor extended-set run · high reasoningPublished margin ±10.7 · Original set 79.1% ±3.4 · 150 reported samples

27.30%

Kimi K2.7 CodeKimiMercor extended-set run · high reasoningPublished margin ±10 · Original set 78.2% ±3.4 · 150 reported samples

26.70%

Gemini 3.5 FlashGoogleMercor extended-set run · high reasoningPublished margin ±9 · Original set 78.5% ±3.5 · 150 reported samples

22.70%

DeepSeek-V4-ProDeepSeekMercor extended-set run · max reasoningPublished margin ±9.3 · Original set 75.4% ±3.5 · 150 reported samples

22.00%

Gemini 3.1 ProGoogleMercor extended-set run · high reasoningPublished margin ±9.7 · Original set 78.4% ±3.3 · 150 reported samples

21.30%

DeepSeek-V3.2DeepSeekMercor extended-set runPublished margin ±5 · Original set 66% ±3.7 · 150 reported samples

10.00%

MiniMax-M2.7MiniMaxMercor extended-set run · high reasoningPublished margin ±5 · Original set 75.3% ±3.5 · 150 reported samples

6.70%

How to read this Mercor evaluation

We captured Mercor's SWE-bench Verified Extended results on October 1, 2026. The table preserves 33 published configurations, their source identifiers, reasoning settings, and disclosed pass@1 values. These are results from Mercor's evaluation setup; we did not rerun them.

The headline uses the 50 Mercor-built tasks. The paired original-set values refer to the separately listed 500-task public benchmark. Missing baselines remain unreported. We keep the extension and original set separate from their upstream benchmark-owner and provider-run tables.

The headline is the source page's selected Pass@1 metric. The task set, harness, tools, and grader belong to this Mercor run and may differ from another evaluator's results.

The row labels retain published error margins and sample counts when supplied. Missing margins and counts remain unreported; a sample count is not used to reconstruct a task denominator. Except where the dedicated source defines an interval, the confidence level is unreported. Equal displayed scores and overlapping margins do not establish statistical ties. Capture time does not establish individual evaluation dates.

The general methodology lists programmatic fail-to-pass and pass-to-pass unit tests (all must pass) and a judge of None. Its run-count label is k = 3. Those notes describe the original public benchmark; extension-specific changes are not established by that reference.

Snapshot

Pass@133 configurationsMercor task extensionDisplay only

Mercor SWE-bench Verified Extended results Mercor APEX benchmark index Mercor methodology Paired Mercor original-set results Source paper Upstream evaluation code Upstream dataset

The published SWE-bench Verified Extended (Mercor) snapshot places DeepSeek-V4.1-Flash first at 82.70%. The third row is 4.00 points behind. The broader top-10 range is 19.40 points, so the table still separates the published systems.

33 configurations are shown for SWE-bench Verified Extended (Mercor). The benchmark falls in the Coding category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. SWE-bench Verified Extended (Mercor) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About SWE-bench Verified Extended (Mercor)

Tasks

50 Mercor tasks; 500 tasks in the separately reported original set

Format

Pass@1

Difficulty

Source-specific evaluation

The selected headline is Pass@1. Source model identifiers, effort settings, published margins, and reported sample counts remain attached to the source configuration. Public-set results and Mercor task extensions occupy separate tables.

SWE-bench Verified Extended evaluation results Public benchmark source

Freshness and provenance

Version

SWE-bench Verified Extended; Mercor Pass@1

Refresh cadence

Reviewed dated source capture

Staleness state

Dated Mercor configurations

Question availability

Upstream task access and terms; only aggregate results mirrored

Dated Mercor configurationsDisplay 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.

Questions

What does SWE-bench Verified Extended (Mercor) measure?

Editing repositories to resolve verified GitHub issues. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Which model leads the published SWE-bench Verified Extended (Mercor) snapshot?

DeepSeek-V4.1-Flash currently leads the published SWE-bench Verified Extended (Mercor) snapshot with 82.70% pass@1. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on SWE-bench Verified Extended (Mercor)?

The October 1, 2026 capture snapshot contains 33 source configurations.

Last updated: October 1, 2026 capture · 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.