AdvancedIF — Mercor evaluation (AdvancedIF (Mercor))

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

Following complex instructions graded against rubric criteria. 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 AdvancedIF (Mercor) — October 15, 2026 capture

We mirror the published pass@1 view for AdvancedIF (Mercor). Gemini 3.1 Pro leads the public snapshot at 86.70%, followed by Gemini 3.6 Flash (85.30%) and Qwen3.8-Max (85.10%). We do not use these results to rank models overall.

Gemini 3.1 Pro

Google

Single-shot agent · high reasoning

Published margin ±1.4 · 6,580 reported samples

86.70%

Gemini 3.6 Flash

Google

Single-shot agent · high reasoning

Published margin ±1.4 · 6,580 reported samples

85.30%

Qwen3.8-Max

Alibaba

Single-shot agent · xhigh reasoning

Published margin ±1.4 · 6,571 reported samples

85.10%

45 configurationsInstruction FollowingDated Mercor configurationsDisplay onlyUpdated October 15, 2026 capture

Pass@1 table (45 configurations)

Pass@1 results for AdvancedIF (Mercor)
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	Gemini 3.1 ProGoogleSingle-shot agent · high reasoningPublished margin ±1.4 · 6,580 reported samples	86.70%	Not reported	Unknown
2	Gemini 3.6 FlashGoogleSingle-shot agent · high reasoningPublished margin ±1.4 · 6,580 reported samples	85.30%	Not reported	Unknown
3	Qwen3.8-MaxAlibabaSingle-shot agent · xhigh reasoningPublished margin ±1.4 · 6,571 reported samples	85.10%	Not reported	Unknown
4	Fable 5.1AnthropicSingle-shot agent · max reasoningPublished margin ±1.5 · 6,494 reported samples	84.20%	Not reported	Unknown
5	DeepSeek-V4.1-FlashDeepSeekSingle-shot agent · max reasoningPublished margin ±1.4 · 6,580 reported samples	84.00%	Not reported	Unknown
6	Sonnet 5.5AnthropicSingle-shot agent · max reasoningPublished margin ±1.5 · 6,553 reported samples	83.40%	Not reported	Unknown
7	Opus 5.5AnthropicSingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	82.80%	Not reported	Unknown
8	Fable 5.1AnthropicSingle-shot agent · high reasoningPublished margin ±1.5 · 6,580 reported samples	82.60%	Not reported	Unknown
9	GPT-6.1 SolOpenAISingle-shot agent · max reasoningPublished margin ±1.6 · 6,557 reported samples	82.40%	Not reported	Unknown
10	GPT-5.6 Sol (Pro)OpenAISingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	82.10%	Not reported	Unknown
11	GPT-6 AstraOpenAISingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	82.00%	Not reported	Unknown
12	GPT-5.5OpenAISingle-shot agent · xhigh reasoningPublished margin ±1.6 · 6,580 reported samples	81.30%	Not reported	Unknown
13	GPT-6 AstraOpenAISingle-shot agent · xhigh reasoningPublished margin ±1.7 · 6,571 reported samples	81.30%	Not reported	Unknown
14	GPT-5.4OpenAISingle-shot agent · xhigh reasoningPublished margin ±1.6 · 6,580 reported samples	81.00%	Not reported	Unknown
15	GPT-6 SolOpenAISingle-shot agent · max reasoningPublished margin ±1.6 · 6,569 reported samples	80.90%	Not reported	Unknown
16	Grok 4.6xAISingle-shot agent · xhigh reasoningPublished margin ±1.6 · 6,578 reported samples	80.90%	Not reported	Unknown
17	Haiku 5.5AnthropicSingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	80.80%	Not reported	Unknown
18	Opus 5AnthropicSingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	80.50%	Not reported	Unknown
19	Grok 4.7xAISingle-shot agent · xhigh reasoningPublished margin ±1.5 · 6,578 reported samples	80.20%	Not reported	Unknown
20	Muse Spark 1.1MetaSingle-shot agent · xhigh reasoningPublished margin ±1.6 · 6,570 reported samples	79.40%	Not reported	Unknown
21	Grok 4.5xAISingle-shot agent · high reasoningPublished margin ±1.6 · 6,580 reported samples	79.30%	Not reported	Unknown
22	GLM-5.3ZhipuSingle-shot agent · max reasoningPublished margin ±1.7 · 6,558 reported samples	78.20%	Not reported	Unknown
23	GPT-5.6 TerraOpenAISingle-shot agent · max reasoningPublished margin ±1.7 · 6,580 reported samples	78.10%	Not reported	Unknown
24	Qwen3.5AlibabaSingle-shot agentPublished margin ±1.6 · 6,580 reported samples	78.10%	Not reported	Unknown
25	InklingThinking MachinesSingle-shot agent · high reasoningPublished margin ±1.6 · 6,580 reported samples	77.60%	Not reported	Unknown
26	GLM-5.2ZhipuSingle-shot agent · max reasoningPublished margin ±1.6 · 6,580 reported samples	77.40%	Not reported	Unknown
27	Gemma 4 31BGoogleSingle-shot agentPublished margin ±1.7 · 6,580 reported samples	77.00%	Not reported	Unknown
28	DeepSeek-V4-FlashDeepSeekSingle-shot agent · max reasoningPublished margin ±1.7 · 6,580 reported samples	76.60%	Not reported	Unknown
29	GPT-5.6 LunaOpenAISingle-shot agent · max reasoningPublished margin ±1.7 · 6,580 reported samples	76.00%	Not reported	Unknown
30	GPT-6 LunaOpenAISingle-shot agent · max reasoningPublished margin ±1.8 · 6,580 reported samples	74.90%	Not reported	Unknown
31	DeepSeek-V4-ProDeepSeekSingle-shot agent · max reasoningPublished margin ±1.7 · 6,535 reported samples	74.70%	Not reported	Unknown
32	Kimi K2.7 CodeKimiSingle-shot agent · high reasoningPublished margin ±1.7 · 6,580 reported samples	73.80%	Not reported	Unknown
33	Opus 4.8AnthropicSingle-shot agent · max reasoningPublished margin ±1.8 · 6,580 reported samples	73.70%	Not reported	Unknown
34	Gemini 3.5 FlashGoogleSingle-shot agent · high reasoningPublished margin ±1.7 · 6,580 reported samples	73.70%	Not reported	Unknown
35	Sonnet 5AnthropicSingle-shot agent · max reasoningPublished margin ±1.8 · 6,557 reported samples	73.30%	Not reported	Unknown
36	Opus 4.7AnthropicSingle-shot agent · max reasoningPublished margin ±1.9 · 6,580 reported samples	73.10%	Not reported	Unknown
37	Sonnet 4.6AnthropicSingle-shot agent · high reasoningPublished margin ±1.9 · 6,580 reported samples	70.80%	Not reported	Unknown
38	Kimi K2 (Thinking)KimiSingle-shot agent · high reasoningPublished margin ±1.8 · 6,580 reported samples	68.50%	Not reported	Unknown
39	GPT-OSS-120BOpenAISingle-shot agent · high reasoningPublished margin ±1.8 · 6,580 reported samples	68.30%	Not reported	Unknown
40	Fable 5AnthropicSingle-shot agent · max reasoningPublished margin ±1.9 · 6,580 reported samples	65.20%	Not reported	Unknown
41	Kimi K3KimiSingle-shot agent · max reasoningPublished margin ±1.8 · 6,557 reported samples	58.70%	Not reported	Unknown
42	MiniMax-M3MiniMaxSingle-shot agent · high reasoningPublished margin ±1.9 · 6,580 reported samples	56.30%	Not reported	Unknown
43	Nemotron 3 UltraNVIDIASingle-shot agent · high reasoningPublished margin ±1.9 · 6,549 reported samples	53.70%	Not reported	Unknown
44	DeepSeek-V3.2DeepSeekSingle-shot agentPublished margin ±2 · 6,576 reported samples	51.30%	Not reported	Unknown
45	MiniMax-M2.7MiniMaxSingle-shot agent · high reasoningPublished margin ±2 · 6,580 reported samples	47.80%	Not reported	Unknown

How to read this Mercor evaluation

We captured Mercor's AdvancedIF results on October 15, 2026. The table preserves 45 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.

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 llm-as-a-judge (lmaaj) over expert-written rubric criteria and a judge of Gemini 3.5 Flash. Its run-count label is k = 4. The linked result page takes precedence where its setup differs, as noted above.

Snapshot

Pass@145 configurationsMercor evaluationDisplay only

Mercor AdvancedIF results Mercor APEX benchmark index Mercor methodology Source paper Upstream dataset

The published AdvancedIF (Mercor) snapshot places Gemini 3.1 Pro first at 86.70%. The third row is 1.60 points behind. The broader top-10 range is 4.60 points, so many of the published results sit in a relatively narrow band.

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

About AdvancedIF (Mercor)

Tasks

1,645 source-reported 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.

AdvancedIF evaluation results Public benchmark source

Freshness and provenance

Version

AdvancedIF; 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 AdvancedIF (Mercor) measure?

Following complex instructions graded against rubric criteria. This table shows Mercor-run configurations for reference and is excluded from model rankings.

Which model leads the published AdvancedIF (Mercor) snapshot?

Gemini 3.1 Pro currently leads the published AdvancedIF (Mercor) snapshot with 86.70% pass@1. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on AdvancedIF (Mercor)?

The October 15, 2026 capture snapshot contains 45 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.