Benchmark Almanac

GIM

GPT-5.4 Pro 1.63 leads since May 2026
Not in model cards

Models solve problems that combine constraint tracking, reasoning, and communication using broadly accessible knowledge. Model judges score answers against task rubrics. An item response model adjusts for problem difficulty and judge effects to estimate ability.

The bank contains 615 public problems and 205 private problems. Comparable ability scores require a calibrated judge and the matching bank version.

Released
May 2026
Kind
Language reasoning
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
Public, plus a private test set
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
A host
Strongest model tested
GPT-5.4 Pro #19 on the capability index
Last model added
April 2026
Scores refreshed
November 18th, 2026

In the authors’ words

As LLM benchmarks saturate, the evaluation community has pursued two strategies to increase difficulty: escalating knowledge demands (GPQA, HLE) or removing knowledge entirely in favor of abstract reasoning (ARC-AGI). The first conflates memorization with capability; the second divorces reasoning from the practical contexts in which it matters. We take a different approach. The Grounded Integration Measure (GIM) is a benchmark of 820 original problems (615 public, 205 private) where difficulty comes from integration; individual problems require coordinating multiple cognitive operations (constraint satisfaction, state tracking, epistemic vigilance, audience calibration) over broadly accessible knowledge, so that reasoning stays grounded in realistic tasks without being gated on specialized expertise. Each problem is an original expert-authored composition, majority with rubric-decomposed scoring. We calibrate a judge-aware continuous response 2-parameter logistic (2PL) IRT model across 53 test-configurations (unique model x thinking-level pairs) and five calibrated judges, using 203,800 epoch-averaged prompt-judge cells derived from >1M raw judge-scored observations, producing robust ability estimates that correctly order test-configurations even when raw accuracy is distorted by errors, missing data, or judge leniency differences. Using this framework, we present a comprehensive leaderboard spanning 22 models and 47 reporting test-configurations, and conduct what is to our knowledge the most extensive published study of how test-time compute trades off against model capability on a fixed benchmark: 11 models swept across 35 test-configurations. We observe that within-family configuration choices, such as thinking budget and quantization, matter as much as model selection, and increasing thinking tokens has diminishing marginal returns.

Abstract of GIM: Evaluating models via tasks that integrate multiple cognitive domains (arXiv:2605.18663, CC0)
21
Entries in its paper, 16 models we track

GPT-5.4 Pro leads with 1.63 among models released by April 2026 GPT-5.4 Pro leads with 1.63 among models released by April 2026

Best score per entry from the benchmark's paper, read on November 18th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.

The newest model with a score here came out in April 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

  1. 1 GPT-5.4 Pro X-High 1.63
  2. 2 Gemini 3.1 Pro High 1.47
  3. 3 Muse Spark X-High 1.29
  4. 4 GPT-5.4 X-High 1.23
  5. 5 Claude Opus 4.7 High 1.09
  6. 6 Claude Opus 4.6 High 1.06
  7. 7 Muse Spark fp8 High 1.02
  8. 8 Claude Sonnet 4.6 High 0.85
  9. 9 GPT-5 High 0.58
  10. 10 Gemma 4 31B IT Medium 0.17
  11. 11 Gemma 4 31B fp4 Medium 0.09
  12. 12 Gemma 4 31B fp8 Medium 0.09
  13. 13 Gemini 2.5 Pro (snapshot unspecified) High 0.01
  14. 14 GPT-5.4 Mini High -0.02
  15. 15 Gemini 3.1 Flash-Lite High -0.03
  16. 16 gpt-oss-120b Medium -0.56
  17. 17 Claude Opus 4 High -0.63
  18. 18 Claude Haiku 4.5 High -0.71
  19. 19 Llama 4 Maverick OFF -1.48
  20. 20 GPT-4o (snapshot unspecified) OFF -1.53
  21. 21 Llama 3.3 70B OFF -1.65

Source: The GIM paper

#EntryScoreSettingDateSource
1 GPT-5.4 Pro 1.63 X-High Mar 2026 Official leaderboard
2 Gemini 3.1 Pro 1.47 High Feb 2026 Official leaderboard
3 Muse Spark 1.29 X-High Apr 2026 Official leaderboard
4 GPT-5.4 1.23 X-High Mar 2026 Official leaderboard
5 Claude Opus 4.7 1.09 High Apr 2026 Official leaderboard
6 Claude Opus 4.6 1.06 High Feb 2026 Official leaderboard
7 Muse Spark fp8 1.02 High Official leaderboard
8 Claude Sonnet 4.6 0.85 High Feb 2026 Official leaderboard
9 GPT-5 0.58 High Aug 2025 Official leaderboard
10 Gemma 4 31B IT 0.17 Medium Apr 2026 Official leaderboard
11 Gemma 4 31B fp4 0.09 Medium Official leaderboard
12 Gemma 4 31B fp8 0.09 Medium Official leaderboard
13 Gemini 2.5 Pro (snapshot unspecified) 0.01 High Official leaderboard
14 GPT-5.4 Mini -0.02 High Mar 2026 Official leaderboard
15 Gemini 3.1 Flash-Lite -0.03 High Mar 2026 Official leaderboard
16 gpt-oss-120b -0.56 Medium Aug 2025 Official leaderboard
17 Claude Opus 4 -0.63 High May 2025 Official leaderboard
18 Claude Haiku 4.5 -0.71 High Oct 2025 Official leaderboard
19 Llama 4 Maverick -1.48 OFF Apr 2025 Official leaderboard
20 GPT-4o (snapshot unspecified) -1.53 OFF Official leaderboard
21 Llama 3.3 70B -1.65 OFF Dec 2024 Official leaderboard
#ModelScoreSettingDateSource
1 GPT-5.4 Pro 1.63 X-High Mar 2026 Official leaderboard
2 Gemini 3.1 Pro 1.47 High Feb 2026 Official leaderboard
3 Muse Spark 1.29 X-High Apr 2026 Official leaderboard
4 GPT-5.4 1.23 X-High Mar 2026 Official leaderboard
5 Claude Opus 4.7 1.09 High Apr 2026 Official leaderboard
6 Claude Opus 4.6 1.06 High Feb 2026 Official leaderboard
7 Claude Sonnet 4.6 0.85 High Feb 2026 Official leaderboard
8 GPT-5 0.58 High Aug 2025 Official leaderboard
9 Gemma 4 31B IT 0.17 Medium Apr 2026 Official leaderboard
10 GPT-5.4 Mini -0.02 High Mar 2026 Official leaderboard
11 Gemini 3.1 Flash-Lite -0.03 High Mar 2026 Official leaderboard
12 gpt-oss-120b -0.56 Medium Aug 2025 Official leaderboard
13 Claude Opus 4 -0.63 High May 2025 Official leaderboard
14 Claude Haiku 4.5 -0.71 High Oct 2025 Official leaderboard
15 Llama 4 Maverick -1.48 OFF Apr 2025 Official leaderboard
16 Llama 3.3 70B -1.65 OFF Dec 2024 Official leaderboard

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.4 Pro
Medium
1.63
Gemini 3.1 Pro
Low Medium
1.47
GPT-5.4
Low Medium
1.23
Claude Opus 4.6
Low Medium High
1.06
Claude Sonnet 4.6
Low Medium High
0.85
GPT-5
Low Medium High
0.58
Gemma 4 31B IT
non-thinkin… Medium
0.17
GPT-5.4 Mini
Low Medium
-0.02
Gemini 3.1 Flash-Lite
Low Medium High
-0.03
Claude Opus 4
Low High
-0.63
Claude Haiku 4.5
Low
-0.71

Source: The GIM paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.4 Pro X-High 1.63
GPT-5.4 Pro High 1.62
GPT-5.4 Pro Medium 1.55
Gemini 3.1 Pro High 1.47
Gemini 3.1 Pro Medium 1.35
Muse Spark X-High 1.29
GPT-5.4 X-High 1.23
GPT-5.4 High 1.21
Claude Opus 4.7 High 1.09
Claude Opus 4.6 High 1.06
GPT-5.4 Medium 1.04
Gemini 3.1 Pro Low 1.03
Claude Sonnet 4.6 High 0.85
Claude Opus 4.6 Medium 0.76
GPT-5.4 Low 0.60
Claude Sonnet 4.6 Medium 0.59
GPT-5 High 0.58
GPT-5 Medium 0.36
Claude Sonnet 4.6 Low 0.20
Gemma 4 31B IT Medium 0.17
Claude Opus 4.6 Low 0.14
GPT-5.4 Mini High -0.02
Gemini 3.1 Flash-Lite High -0.03
GPT-5 Low -0.04
GPT-5.4 Mini Medium -0.13
gpt-oss-120b Medium -0.56
Gemini 3.1 Flash-Lite Medium -0.63
Claude Opus 4 High -0.63
Claude Haiku 4.5 High -0.71
GPT-5.4 Mini Low -0.76
Claude Haiku 4.5 Medium -0.81
Claude Opus 4 Medium -0.83
Gemma 4 31B IT non-thinking, OFF -0.85
Gemini 3.1 Flash-Lite Low -0.91
Claude Haiku 4.5 Low -0.91
Claude Opus 4 Low -0.92
Llama 4 Maverick OFF -1.48
Llama 3.3 70B OFF -1.65

Scores at each effort level

Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.

Lowest effort to highest, light to dark
GPT-5.4 Pro
Medium
1.63
Gemini 3.1 Pro
Low Medium
1.47
GPT-5.4
Low Medium
1.23
Claude Opus 4.6
Low Medium High
1.06
Claude Sonnet 4.6
Low Medium High
0.85
GPT-5
Low Medium High
0.58
Gemma 4 31B IT
non-thinkin… Medium
0.17
GPT-5.4 Mini
Low Medium
-0.02
Gemini 3.1 Flash-Lite
Low Medium High
-0.03
Claude Opus 4
Low High
-0.63
Claude Haiku 4.5
Low
-0.71

Source: The GIM paper

ModelEffortHarnessScoreCost per taskOutput tokens
GPT-5.4 Pro X-High 1.63
GPT-5.4 Pro High 1.62
GPT-5.4 Pro Medium 1.55
Gemini 3.1 Pro High 1.47
Gemini 3.1 Pro Medium 1.35
Muse Spark X-High 1.29
GPT-5.4 X-High 1.23
GPT-5.4 High 1.21
Claude Opus 4.7 High 1.09
Claude Opus 4.6 High 1.06
GPT-5.4 Medium 1.04
Gemini 3.1 Pro Low 1.03
Claude Sonnet 4.6 High 0.85
Claude Opus 4.6 Medium 0.76
GPT-5.4 Low 0.60
Claude Sonnet 4.6 Medium 0.59
GPT-5 High 0.58
GPT-5 Medium 0.36
Claude Sonnet 4.6 Low 0.20
Gemma 4 31B IT Medium 0.17
Claude Opus 4.6 Low 0.14
GPT-5.4 Mini High -0.02
Gemini 3.1 Flash-Lite High -0.03
GPT-5 Low -0.04
GPT-5.4 Mini Medium -0.13
gpt-oss-120b Medium -0.56
Gemini 3.1 Flash-Lite Medium -0.63
Claude Opus 4 High -0.63
Claude Haiku 4.5 High -0.71
GPT-5.4 Mini Low -0.76
Claude Haiku 4.5 Medium -0.81
Claude Opus 4 Medium -0.83
Gemma 4 31B IT non-thinking, OFF -0.85
Gemini 3.1 Flash-Lite Low -0.91
Claude Haiku 4.5 Low -0.91
Claude Opus 4 Low -0.92
Llama 4 Maverick OFF -1.48
Llama 3.3 70B OFF -1.65

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Benchmark paper Not tested since May 2026

Source: The GIM paper

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Plot with other benchmarks
In the benchmark’s paper
-2.00 -1.00 0.00 1.00 2.00 2025 2026 -2.00 -1.00 0.00 1.00 2.00 2025 2026

Source: Official GIM leaderboard

No lab model card reports it yet

We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.

More in Knowledge

All 332 in Knowledge
  • MMLU-Pro In use Harder 10-option version of MMLU with reasoning-focused questions across 14 subjects. Intern-S2-397B 89.8%
  • MMLU In use 57-subject multiple-choice exam questions from elementary to professional level. Scored by accuracy. Seed1.8 92.3%
  • SuperGPQA In use 26,529 graduate-level multiple-choice questions across 285 disciplines. Accuracy. Qwen 3.6 Max (Preview) 73.9%
  • WideSearch In use 200 bilingual broad info-seeking tasks: agent collects many atomic facts into a table. Scored by success rate and row/item F1. Tencent Hy4 preview 83.9%
  • BBEH micro average In use Models solve harder replacements for each of the 23 BIG-Bench Hard tasks, 4,520 examples in total. The score is the share of all examples answered correctly. Gemma 4 31B IT 74.4%
  • HLE-Diamond In use Models answer 1,000 cleaned expert questions from Humanity's Last Exam, 500 on reasoning and 500 on knowledge, closed book. Graded on exact answers. Claude Fable 5.2 77.9%

Data audit

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for Claude Opus 4.7. Every benchmark

Catalog record 2 pass
  • Pass Task access Public tasks, plus a private test set.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the benchmark’s paper (21 entries).
  • Pass Models scored 16 models we track, plus 5 other entries on its leaderboard.
  • Warning Newest score The newest came 7 months ago, in April 2026, for Claude Opus 4.7.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 Pro (#19).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 38 rows for models we track and 5 other entries at the last read.
  • Pass Last read November 18th, 2026.
  • Pass Read worked The last read worked.
Consistency 1 pass 1 note
  • Note Lab claims No model has both a lab number and an independent run.
  • Pass Known issues No published audit found problems.

How we check