APEX-SWE (Inspect) — Mercor evaluation (APEX-SWE (Inspect) (Mercor))

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

Software integration and production-failure diagnosis with the Inspect harness. 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 APEX-SWE (Inspect) (Mercor) — October 15, 2026 capture

We mirror the published pass@1 view for APEX-SWE (Inspect) (Mercor). Fable 5 leads the public snapshot at 54.83%, followed by Opus 5 (54.67%) and Grok 4.5 (51.17%). We do not use these results to rank models overall.

Fable 5

Anthropic

Inspect · max reasoning

Published margin ±6 · 799 reported samples

54.83%

Opus 5

Anthropic

Inspect · max reasoning

Published margin ±5.5 · 800 reported samples

54.67%

Grok 4.5

xAI

Inspect · high reasoning

Published margin ±6 · 800 reported samples

51.17%

23 configurationsCodingDated Mercor configurationsDisplay onlyUpdated October 15, 2026 capture

Pass@1 table (23 configurations)

Pass@1 results for APEX-SWE (Inspect) (Mercor)
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Fable 5AnthropicInspect · max reasoningPublished margin ±6 · 799 reported samples	54.83%	Not reported	Unknown
2	Opus 5AnthropicInspect · max reasoningPublished margin ±5.5 · 800 reported samples	54.67%	Not reported	Unknown
3	Grok 4.5xAIInspect · high reasoningPublished margin ±6 · 800 reported samples	51.17%	Not reported	Unknown
4	Opus 4.8AnthropicInspect · high reasoningPublished margin ±6 · Sample count not reported	47.30%	Not reported	Unknown
5	Kimi K3KimiInspect · max reasoningPublished margin ±6 · 800 reported samples	44.67%	Not reported	Unknown
6	Sonnet 5AnthropicInspect · high reasoningPublished margin ±6 · Sample count not reported	43.65%	Not reported	Unknown
7	GPT-5.3-CodexOpenAIInspect · high reasoningPublished margin ±6.25 · Sample count not reported	41.50%	Not reported	Unknown
8	GPT-5.5OpenAIInspect · xhigh reasoningPublished margin ±6.5 · 1,600 reported samples	40.84%	Not reported	Unknown
9	GPT-5.6 SolOpenAIInspect · xhigh reasoningPublished margin ±5.5 · 800 reported samples	39.67%	Not reported	Unknown
10	Gemini 3.6 FlashGoogleInspect · high reasoningPublished margin ±6 · 800 reported samples	39.67%	Not reported	Unknown
11	GLM-5.2ZhipuInspectPublished margin ±5.5 · Sample count not reported	37.32%	Not reported	Unknown
12	Gemini 3.5 FlashGoogleInspect · high reasoningPublished margin ±5.8 · 800 reported samples	32.50%	Not reported	Unknown
13	Kimi K2.7KimiInspectPublished margin ±6 · Sample count not reported	32.48%	Not reported	Unknown
14	Cognition SWE-1.6 PreviewCognitionInspectPublished margin ±5.75 · Sample count not reported	31.70%	Not reported	Unknown
15	Sonnet 4.5AnthropicInspect · high reasoningPublished margin ±6 · Sample count not reported	31.00%	Not reported	Unknown
16	GPT-5.2-CodexOpenAIInspect · high reasoningPublished margin ±5.84 · Sample count not reported	30.80%	Not reported	Unknown
17	Kimi K2.5KimiInspectPublished margin ±5.83 · Sample count not reported	30.30%	Not reported	Unknown
18	GPT-5.1-CodexOpenAIInspect · high reasoningPublished margin ±5.59 · Sample count not reported	29.00%	Not reported	Unknown
19	Gemini 3.1 ProGoogleInspect · high reasoningPublished margin ±5.42 · 800 reported samples	28.80%	Not reported	Unknown
20	Gemini 3 ProGoogleInspect · high reasoningPublished margin ±5.5 · Sample count not reported	26.80%	Not reported	Unknown
21	Grok 4xAIInspectPublished margin ±5 · Sample count not reported	21.00%	Not reported	Unknown
22	DeepSeek-V3.2DeepSeekInspectPublished margin ±4.33 · 800 reported samples	15.80%	Not reported	Unknown
23	Kimi K2 (Thinking)KimiInspectPublished margin ±3.58 · Sample count not reported	11.20%	Not reported	Unknown

How to read this Mercor evaluation

We captured Mercor's APEX-SWE (Inspect) results on October 15, 2026. The table preserves 23 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.

This dated source table is display only and does not enter overall or category rankings. Source configurations retain their external identity, including Pro descriptors and different effort settings.

These are the disclosed Inspect Pass@1 results across integration and observability work. Terminus-2 is a separate harness profile. One source record labels its metric pass@1 rather than the declared pass-1 selector; it is retained in private evidence and withheld from this table until the source resolves that mismatch.

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 expert-authored rubrics plus repository test suites and a judge of LLM judge + programmatic tests. Its run-count label is k = 3. The linked result page takes precedence where its setup differs, as noted above.

Snapshot

Pass@123 configurationsMercor evaluationDisplay only

Mercor APEX-SWE (Inspect) results Mercor APEX benchmark index Mercor methodology Source paper Upstream evaluation code Upstream dataset Mercor evaluation notes

The published APEX-SWE (Inspect) (Mercor) snapshot places Fable 5 first at 54.83%. The third row is 3.66 points behind. The broader top-10 range is 15.16 points, so the table still separates the published systems.

23 configurations are shown for APEX-SWE (Inspect) (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. APEX-SWE (Inspect) (Mercor) is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About APEX-SWE (Inspect) (Mercor)

Tasks

Integration and observability engineering tasks

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.

APEX-SWE (Inspect) evaluation results Public benchmark source

Freshness and provenance

Version

APEX-SWE (Inspect); Mercor Pass@1; inspect

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 APEX-SWE (Inspect) (Mercor) measure?

Software integration and production-failure diagnosis with the Inspect harness. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Which model leads the published APEX-SWE (Inspect) (Mercor) snapshot?

Fable 5 currently leads the published APEX-SWE (Inspect) (Mercor) snapshot with 54.83% pass@1. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on APEX-SWE (Inspect) (Mercor)?

The October 15, 2026 capture snapshot contains 23 source configurations.

Last updated: October 15, 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 5,600+ readers.

One email each week. Unsubscribe anytime.