Benchmark Almanac

PerspectiveGap

GPT-5.5 62% leads since June 2026
Not in model cards

Models assign information fragments to workers with different roles. A strict pass requires supplying all needed information without disclosing extra fragments.

Made by
Youran Sun and colleagues
Released
June 2026
Kind
Following precise instructions
Level
University
Scored
Checked automatically
Answers
Short answer
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
35
Models scored on its own leaderboard

GPT-5.5 leads with 62% among models released by September 2026

Best score per model from the benchmark's own leaderboard, read on November 20th, 2026. Where a model ran with several settings or harnesses, we keep its best.

The newest model with a score here came out in September 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-5.5 62%
  2. 2 GPT-5.6 Terra 42.7%
  3. 3 GPT-5.6 Sol 42.5%
  4. 4 Grok 4.6 40.9%
  5. 5 DeepSeek-V4-Pro 32%
  6. 6 Claude Fable 5 31.4%
  7. 7 Kimi K2.6 25.9%
  8. 8 Claude Sonnet 5 25.7%
  9. 9 DeepSeek-V4-Flash 24.8%
  10. 10 Gemini 3.1 Pro 21.8%
  11. 11 GLM-5 20.2%
  12. 12 MiMo-V2.5-Pro 20%
  13. 13 Claude Opus 4.7 19.1%
  14. 14 Qwen 3.6 Plus 17%
  15. 15 GPT-5.6 Luna 16.6%
  16. 16 Claude Sonnet 4.6 15.7%
  17. 17 Grok 4.3 15.5%
  18. 18 Kimi K2.5 15.2%
  19. 19 Qwen3.7-Max 14.3%
  20. 20 GPT-5.4 13.9%
  21. 21 Claude Opus 4.8 13.9%
  22. 22 Qwen3.5-9B 12.7%
  23. 23 MiMo-V2.5 12.7%
  24. 24 MiniMax-M2.7 11.1%
  25. 25 Gemini 3.5 Flash 10%
  26. 26 GLM-5.1 9.50%
  27. 27 GPT-6 Astra 8.90%
  28. 28 Claude Haiku 4.5 5.70%
  29. 29 MiniMax-M2.5 5.50%
  30. 30 Grok 4.20 5.00%
  31. 31 glm-4.7-flash 3.40%
  32. 32 Gemini 3.1 Flash-Lite 3.40%
  33. 33 MiMo-V2-Flash 2.70%
  34. 34 GPT-5.4 Mini 1.80%
  35. 35 GPT-5.4 Nano 0.50%

Source: Official PerspectiveGap leaderboard

#ModelScoreSettingDateSource
1 GPT-5.5 62% Apr 2026 Official leaderboard
2 GPT-5.6 Terra 42.7% Jul 2026 Official leaderboard
3 GPT-5.6 Sol 42.5% Jul 2026 Official leaderboard
4 Grok 4.6 40.9% Aug 2026 Official leaderboard
5 DeepSeek-V4-Pro 32% Apr 2026 Official leaderboard
6 Claude Fable 5 31.4% Jun 2026 Official leaderboard
7 Kimi K2.6 25.9% Apr 2026 Official leaderboard
8 Claude Sonnet 5 25.7% Jun 2026 Official leaderboard
9 DeepSeek-V4-Flash 24.8% Apr 2026 Official leaderboard
10 Gemini 3.1 Pro 21.8% Feb 2026 Official leaderboard
11 GLM-5 20.2% Feb 2026 Official leaderboard
12 MiMo-V2.5-Pro 20% Apr 2026 Official leaderboard
13 Claude Opus 4.7 19.1% Apr 2026 Official leaderboard
14 Qwen 3.6 Plus 17% Mar 2026 Official leaderboard
15 GPT-5.6 Luna 16.6% Jul 2026 Official leaderboard
16 Claude Sonnet 4.6 15.7% Feb 2026 Official leaderboard
17 Grok 4.3 15.5% Apr 2026 Official leaderboard
18 Kimi K2.5 15.2% Jan 2026 Official leaderboard
19 Qwen3.7-Max 14.3% May 2026 Official leaderboard
20 GPT-5.4 13.9% Mar 2026 Official leaderboard
21 Claude Opus 4.8 13.9% May 2026 Official leaderboard
22 Qwen3.5-9B 12.7% Feb 2026 Official leaderboard
23 MiMo-V2.5 12.7% Apr 2026 Official leaderboard
24 MiniMax-M2.7 11.1% Mar 2026 Official leaderboard
25 Gemini 3.5 Flash 10% May 2026 Official leaderboard
26 GLM-5.1 9.50% Apr 2026 Official leaderboard
27 GPT-6 Astra 8.90% Sep 2026 Official leaderboard
28 Claude Haiku 4.5 5.70% Oct 2025 Official leaderboard
29 MiniMax-M2.5 5.50% Feb 2026 Official leaderboard
30 Grok 4.20 5.00% Feb 2026 Official leaderboard
31 glm-4.7-flash 3.40% Jan 2026 Official leaderboard
32 Gemini 3.1 Flash-Lite 3.40% Mar 2026 Official leaderboard
33 MiMo-V2-Flash 2.70% Dec 2025 Official leaderboard
34 GPT-5.4 Mini 1.80% Mar 2026 Official leaderboard
35 GPT-5.4 Nano 0.50% Mar 2026 Official leaderboard

Rankings by part

We read 12 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Role-fragment assignment, Prompt writing and coder reviewer. GPT-5.5 leads every one of them. Arrows show how far a model moved from its overall rank among the models we track.

Role-fragment assignment

35 models we track have a score here.

  1. 1 GPT-5.5 55.5%
  2. 2 GPT-5.6 Terra 43.6%
  3. 3 Grok 4.6 ↑1 43.2%
  4. 4 Claude Fable 5 ↑2 38.6%
  5. 5 DeepSeek-V4-Pro 37.3%
  6. 6 GPT-5.6 Sol ↓3 35.9%
  7. 7 Claude Sonnet 5 ↑1 35%
  8. 8 Kimi K2.6 ↓1 26.8%
  9. 9 GLM-5 ↑2 25.9%
  10. 10 GPT-5.4 ↑10 25.5%

Number one over time

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

OpenAI
Own leaderboard Not tested since Sep 2026
GPT-5.5 62%

Source: Official PerspectiveGap leaderboard

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
On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Kimi K2.6 GLM-5 Kimi K2.5 Claude Haiku 4.5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-5.5 Kimi K2.6

Source: Official PerspectiveGap 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 Instructions and chat

All 144 in Instructions and chat
  • IFEval In use Models answer about 500 prompts containing verifiable instructions such as word limits or required keywords. Code checks each instruction for strict and loose accuracy. Gemma 4 31B IT 98.9%
  • IFBench In use 58 new out-of-domain verifiable output constraints. Scored by programmatic constraint checks (strict or loose). Flint 89.2%
  • MultiChallenge In use Multi-turn conversations testing instruction retention, inference memory, versioned editing and self-coherence. Graded by LLM judge with rubrics. GPT-6.1 Sol 82%
  • Multi-IF In use Models follow verifiable instructions over three-turn conversations in eight languages, extending IFEval. Scored by instruction accuracy at each turn. LFM3-1.7B 89.1%
  • AdvancedIF In use Models respond to over 1,600 prompts with complex, multi-turn, and system-level instructions. A judge model checks each response against expert-written rubrics. MAI-Thinking-1 85%
  • MM-IFEval In use Multimodal models follow image-grounded instructions with exact output constraints. Checks mix rule-based verification and a judge model. GPT-4o 64.6%

Data audit

All 10 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 4 pass 1 note
  • Pass Independent runs Scores from the official leaderboard (35 entries).
  • Pass Models scored 35 models we track.
  • Pass Newest score The newest is from September 2026, for GPT-6 Astra.
  • Note Frontier models Only 1 of the top 10 has a score. The strongest tested is GPT-6 Astra (#4).
  • Pass Headroom The best is 62%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 35 rows for models we track at the last read.
  • Pass Last read November 20th, 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