PerspectiveGap
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
- 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
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:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- Claude Opus 5.5
- and 12 more
- 62.0 GPT-5.5
- 42.7 GPT-5.6 Terra
- 42.5 GPT-5.6 Sol
- 40.9 Grok 4.6
- 32.0
DeepSeek-V4-Pro
- 31.4 Claude Fable 5
- 25.9 Kimi K2.6
- 25.7 Claude Sonnet 5
- 24.8
DeepSeek-V4-Flash
- 21.8
Gemini 3.1 Pro
- 20.2 GLM-5
- 20.0 MiMo-V2.5-Pro
- 19.1 Claude Opus 4.7
- 17.0
Qwen 3.6 Plus
- 16.6 GPT-5.6 Luna
- 15.7 Claude Sonnet 4.6
- 15.5 Grok 4.3
- 15.2 Kimi K2.5
- 14.3
Qwen3.7-Max
- 13.9 GPT-5.4
- 13.9 Claude Opus 4.8
- 12.7
Qwen3.5-9B
- 12.7 MiMo-V2.5
- 11.1
MiniMax-M2.7
- 10.0
Gemini 3.5 Flash
- 9.50 GLM-5.1
- 8.90 GPT-6 Astra
- 5.70 Claude Haiku 4.5
- 5.50
MiniMax-M2.5
- 5.00 Grok 4.20
- 3.40 glm-4.7-flash
- 3.40
Gemini 3.1 Flash-Lite
- 2.70 MiMo-V2-Flash
- 1.80 GPT-5.4 Mini
- 0.50 GPT-5.4 Nano
- 1 GPT-5.5 62%
- 2 GPT-5.6 Terra 42.7%
- 3 GPT-5.6 Sol 42.5%
- 4 Grok 4.6 40.9%
- 5
DeepSeek-V4-Pro 32%
- 6 Claude Fable 5 31.4%
- 7 Kimi K2.6 25.9%
- 8 Claude Sonnet 5 25.7%
- 9
DeepSeek-V4-Flash 24.8%
- 10
Gemini 3.1 Pro 21.8%
- 11 GLM-5 20.2%
- 12 MiMo-V2.5-Pro 20%
- 13 Claude Opus 4.7 19.1%
- 14
Qwen 3.6 Plus 17%
- 15 GPT-5.6 Luna 16.6%
- 16 Claude Sonnet 4.6 15.7%
- 17 Grok 4.3 15.5%
- 18 Kimi K2.5 15.2%
- 19
Qwen3.7-Max 14.3%
- 20 GPT-5.4 13.9%
- 21 Claude Opus 4.8 13.9%
- 22
Qwen3.5-9B 12.7%
- 23 MiMo-V2.5 12.7%
- 24
MiniMax-M2.7 11.1%
- 25
Gemini 3.5 Flash 10%
- 26 GLM-5.1 9.50%
- 27 GPT-6 Astra 8.90%
- 28 Claude Haiku 4.5 5.70%
- 29
MiniMax-M2.5 5.50%
- 30 Grok 4.20 5.00%
- 31 glm-4.7-flash 3.40%
- 32
Gemini 3.1 Flash-Lite 3.40%
- 33 MiMo-V2-Flash 2.70%
- 34 GPT-5.4 Mini 1.80%
- 35 GPT-5.4 Nano 0.50%
Source: Official PerspectiveGap leaderboard
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 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 | | 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 | | 24.8% | Apr 2026 | Official leaderboard | |
| 10 | | 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 | | 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 | | 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 | | 12.7% | Feb 2026 | Official leaderboard | |
| 23 | MiMo-V2.5 | 12.7% | Apr 2026 | Official leaderboard | |
| 24 | | 11.1% | Mar 2026 | Official leaderboard | |
| 25 | | 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 | | 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 | | 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 GPT-5.5 55.5%
- 2 GPT-5.6 Terra 43.6%
- 3 Grok 4.6 ↑1 43.2%
- 4 Claude Fable 5 ↑2 38.6%
- 5
DeepSeek-V4-Pro 37.3%
- 6 GPT-5.6 Sol ↓3 35.9%
- 7 Claude Sonnet 5 ↑1 35%
- 8 Kimi K2.6 ↓1 26.8%
- 9 GLM-5 ↑2 25.9%
- 10 GPT-5.4 ↑10 25.5%
Prompt writing
35 models we track have a score here.
- 1 GPT-5.5 68.6%
- 2 GPT-5.6 Sol ↑1 49.1%
- 3 GPT-5.6 Terra ↓1 41.8%
- 4 Grok 4.6 38.6%
- 5
DeepSeek-V4-Pro 26.8%
- 6 Claude Sonnet 4.6 ↑10 25.5%
- 7 Kimi K2.6 25%
- 8
DeepSeek-V4-Flash ↑1 25%
- 9 Claude Fable 5 ↓3 24.1%
- 10
Gemini 3.1 Pro 23.6%
coder reviewer
12 models we track have a score here.
- 1 GPT-5.5 84.1%
- 2
DeepSeek-V4-Flash ↑7 52.3%
- 3
DeepSeek-V4-Pro ↑2 47.7%
- 4 MiMo-V2.5-Pro ↑8 45.5%
- 5 GLM-5 ↑6 36.4%
- 6 Kimi K2.6 ↑1 20.5%
- 7 Kimi K2.5 ↑11 15.9%
- 8
Qwen 3.6 Plus ↑6 15.9%
- 9 Claude Opus 4.7 ↑4 11.4%
- 10 Grok 4.3 ↑7 11.4%
coder reviewer pool
12 models we track have a score here.
- 1 GPT-5.5 84.1%
- 2
DeepSeek-V4-Flash ↑7 70.5%
- 3 Grok 4.3 ↑14 63.6%
- 4
DeepSeek-V4-Pro ↑1 50%
- 5
Gemini 3.1 Pro ↑5 47.7%
- 6 MiMo-V2.5-Pro ↑6 47.7%
- 7 Kimi K2.6 40.9%
- 8 Kimi K2.5 ↑10 34.1%
- 9 GLM-5 ↑2 34.1%
- 10 Claude Opus 4.7 ↑3 31.8%
dispatcher coder reviewer
12 models we track have a score here.
- 1 GPT-5.5 86.4%
- 2
Gemini 3.1 Pro ↑8 45.5%
- 3
DeepSeek-V4-Pro ↑2 45.5%
- 4 Claude Opus 4.7 ↑9 38.6%
- 5 Kimi K2.6 ↑2 38.6%
- 6 Claude Sonnet 4.6 ↑10 36.4%
- 7 MiMo-V2.5-Pro ↑5 31.8%
- 8
Qwen 3.6 Plus ↑6 29.5%
- 9
DeepSeek-V4-Flash 25%
- 10 Kimi K2.5 ↑8 18.2%
dispatcher coder reviewer pool
12 models we track have a score here.
- 1 GPT-5.5 75%
- 2
DeepSeek-V4-Pro ↑3 52.3%
- 3
Gemini 3.1 Pro ↑7 47.7%
- 4
DeepSeek-V4-Flash ↑5 34.1%
- 5 Claude Sonnet 4.6 ↑11 31.8%
- 6
Qwen 3.6 Plus ↑8 31.8%
- 7 GLM-5 ↑4 25%
- 8 Kimi K2.6 ↓1 20.5%
- 9 Kimi K2.5 ↑9 15.9%
- 10 Grok 4.3 ↑7 13.6%
dispatcher planloop codeloop
12 models we track have a score here.
- 1 GPT-5.5 38.6%
- 2 Claude Opus 4.7 ↑11 20.5%
- 3 Kimi K2.6 ↑4 20.5%
- 4 Claude Sonnet 4.6 ↑12 9.10%
- 5
DeepSeek-V4-Pro 9.10%
- 6 GLM-5 ↑5 6.80%
- 7 Kimi K2.5 ↑11 2.30%
- 8
Gemini 3.1 Pro ↑2 2.30%
- 9 Grok 4.3 ↑8 2.30%
- 10
Qwen 3.6 Plus ↑4 0%
dispatcher planloop codeloop pool
12 models we track have a score here.
- 1 GPT-5.5 45.5%
- 2
Gemini 3.1 Pro ↑8 13.6%
- 3
DeepSeek-V4-Pro ↑2 13.6%
- 4 Kimi K2.6 ↑3 9.10%
- 5 Claude Sonnet 4.6 ↑11 6.80%
- 6
Qwen 3.6 Plus ↑8 6.80%
- 7 Claude Opus 4.7 ↑6 4.50%
- 8 Kimi K2.5 ↑10 0%
- 9 GLM-5 ↑2 0%
- 10 MiMo-V2.5-Pro ↑2 0%
dispatcher scientist coder reviewer
12 models we track have a score here.
- 1 GPT-5.5 88.6%
- 2 Kimi K2.6 ↑5 79.5%
- 3
DeepSeek-V4-Pro ↑2 75%
- 4 Claude Opus 4.7 ↑9 68.2%
- 5 Kimi K2.5 ↑13 59.1%
- 6 GLM-5 ↑5 52.3%
- 7 MiMo-V2.5-Pro ↑5 50%
- 8
DeepSeek-V4-Flash ↑1 50%
- 9
Gemini 3.1 Pro ↑1 47.7%
- 10
Qwen 3.6 Plus ↑4 43.2%
dispatcher theoryloop codeloop
12 models we track have a score here.
- 1 GPT-5.5 27.3%
- 2 Kimi K2.6 ↑5 13.6%
- 3 GLM-5 ↑8 6.80%
- 4 Claude Sonnet 4.6 ↑12 2.30%
- 5
Gemini 3.1 Pro ↑5 2.30%
- 6 Claude Opus 4.7 ↑7 2.30%
- 7 Kimi K2.5 ↑11 0%
- 8
Qwen 3.6 Plus ↑6 0%
- 9 MiMo-V2.5-Pro ↑3 0%
- 10
DeepSeek-V4-Flash ↓1 0%
dispatcher theoryloop codeloop pool
12 models we track have a score here.
- 1 GPT-5.5 13.6%
- 2
DeepSeek-V4-Pro ↑3 9.10%
- 3 Kimi K2.6 ↑4 4.50%
- 4 Kimi K2.5 ↑14 2.30%
- 5
Gemini 3.1 Pro ↑5 2.30%
- 6
Qwen 3.6 Plus ↑8 2.30%
- 7 MiMo-V2.5-Pro ↑5 2.30%
- 8 GLM-5 ↑3 0%
- 9 Claude Sonnet 4.6 ↑7 0%
- 10 Claude Opus 4.7 ↑3 0%
supervisor phdstudent librarian
12 models we track have a score here.
- 1 GPT-5.5 77.3%
- 2 GLM-5 ↑9 22.7%
- 3
DeepSeek-V4-Pro ↑2 18.2%
- 4
DeepSeek-V4-Flash ↑5 15.9%
- 5 Claude Sonnet 4.6 ↑11 13.6%
- 6
Qwen 3.6 Plus ↑8 11.4%
- 7 Kimi K2.6 11.4%
- 8 MiMo-V2.5-Pro ↑4 11.4%
- 9 Grok 4.3 ↑8 9.10%
- 10 Kimi K2.5 ↑8 4.50%
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
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.
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
- 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.