Benchmark Almanac

Hanabi Agent Benchmark

Claude Opus 5.5 20.5 leads since October 2026
Not in model cards

Models play cooperative Hanabi in four-player teams. The benchmark scores shared board progress while hints and lives constrain coordination.

The October 2026 dataset compares 21 baseline model configurations. Claude Opus 5.5 leads with mean progress 20.50 out of 25 across four seeds.

Made by
flatkey
Released
October 2026
Kind
Multi-agent games
Level
Everyday
Scored
Settled by events
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Actions
Interaction
Multi-agent
Environment
Hanabi
Items from
Live play or votes
Refresh
Live
Run by
Anyone
In views
Games, Agents, Kept fresh
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
October 2026
Scores refreshed
November 20th, 2026
21
Entries on its own leaderboard, 19 models we track

Claude Opus 5.5 leads with 20.5 Claude Opus 5.5 leads with 20.5

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

  1. 1 Claude Opus 5.5 Low 20.5
  2. 2 GPT-6 Astra Low 18.8
  3. 3 Gemini 3.1 Pro Low 18.0
  4. 4 Claude Fable 5.1 Low 18.0
  5. 5 GPT-6 Sol Low 16.3
  6. 6 Qwen 3.8 Max Single 16.0
  7. 7 Gemini 3.8 Flash Low 16.0
  8. 8 Seed 2.1 Lite Single 15.5
  9. 9 DeepSeek-V4-Pro Single 15.3
  10. 10 DeepSeek Flash Low 14.3
  11. 11 Tencent Hy3 Low 13.3
  12. 12 Claude Haiku 5.5 Low 12.8
  13. 13 Seed2.1 Pro Single 12.5
  14. 14 Kimi K3 Low 11.3
  15. 15 Claude Sonnet 5.5 Low 10.5
  16. 16 Qwen3.8 Flash Single 9.50
  17. 17 MiMo-V2.6 Flash Single 9.50
  18. 18 MiMo-V2.6 Pro Single 7.25
  19. 19 GPT-6 Luna Low 6.75
  20. 20 GLM-5.3-Flash Low 5.75
  21. 21 GLM-5.3 Low 4.50

Source: Official Hanabi Agent Benchmark leaderboard

#EntryScoreSettingDateSource
1 Claude Opus 5.5 20.5 Low Sep 2026 Official leaderboard
2 GPT-6 Astra 18.8 Low Sep 2026 Official leaderboard
3 Gemini 3.1 Pro 18.0 Low Feb 2026 Official leaderboard
4 Claude Fable 5.1 18.0 Low Sep 2026 Official leaderboard
5 GPT-6 Sol 16.3 Low Sep 2026 Official leaderboard
6 Qwen 3.8 Max 16.0 Single Aug 2026 Official leaderboard
7 Gemini 3.8 Flash 16.0 Low Sep 2026 Official leaderboard
8 Seed 2.1 Lite 15.5 Single Official leaderboard
9 DeepSeek-V4-Pro 15.3 Single Apr 2026 Official leaderboard
10 DeepSeek Flash 14.3 Low Official leaderboard
11 Tencent Hy3 13.3 Low Jul 2026 Official leaderboard
12 Claude Haiku 5.5 12.8 Low Oct 2026 Official leaderboard
13 Seed2.1 Pro 12.5 Single Jun 2026 Official leaderboard
14 Kimi K3 11.3 Low Jul 2026 Official leaderboard
15 Claude Sonnet 5.5 10.5 Low Sep 2026 Official leaderboard
16 Qwen3.8 Flash 9.50 Single Aug 2026 Official leaderboard
17 MiMo-V2.6 Flash 9.50 Single Sep 2026 Official leaderboard
18 MiMo-V2.6 Pro 7.25 Single Sep 2026 Official leaderboard
19 GPT-6 Luna 6.75 Low Sep 2026 Official leaderboard
20 GLM-5.3-Flash 5.75 Low Aug 2026 Official leaderboard
21 GLM-5.3 4.50 Low Aug 2026 Official leaderboard
#ModelScoreSettingDateSource
1 Claude Opus 5.5 20.5 Low Sep 2026 Official leaderboard
2 GPT-6 Astra 18.8 Low Sep 2026 Official leaderboard
3 Gemini 3.1 Pro 18.0 Low Feb 2026 Official leaderboard
4 Claude Fable 5.1 18.0 Low Sep 2026 Official leaderboard
5 GPT-6 Sol 16.3 Low Sep 2026 Official leaderboard
6 Qwen 3.8 Max 16.0 Single Aug 2026 Official leaderboard
7 Gemini 3.8 Flash 16.0 Low Sep 2026 Official leaderboard
8 DeepSeek-V4-Pro 15.3 Single Apr 2026 Official leaderboard
9 Tencent Hy3 13.3 Low Jul 2026 Official leaderboard
10 Claude Haiku 5.5 12.8 Low Oct 2026 Official leaderboard
11 Seed2.1 Pro 12.5 Single Jun 2026 Official leaderboard
12 Kimi K3 11.3 Low Jul 2026 Official leaderboard
13 Claude Sonnet 5.5 10.5 Low Sep 2026 Official leaderboard
14 Qwen3.8 Flash 9.50 Single Aug 2026 Official leaderboard
15 MiMo-V2.6 Flash 9.50 Single Sep 2026 Official leaderboard
16 MiMo-V2.6 Pro 7.25 Single Sep 2026 Official leaderboard
17 GPT-6 Luna 6.75 Low Sep 2026 Official leaderboard
18 GLM-5.3-Flash 5.75 Low Aug 2026 Official leaderboard
19 GLM-5.3 4.50 Low Aug 2026 Official leaderboard

Rankings by part

We read 4 parts from the benchmark’s own leaderboard, each ranking models on one part of the test, such as Seed 101, Seed 102 and Seed 103. Claude Opus 5.5 leads 3 of them, and Claude Fable 5.1 leads the other one. Arrows show how far a model moved from its overall rank among the models we track.

Seed 101

19 models we track have a score here.

  1. 1 Claude Opus 5.5 20.0
  2. 2 Claude Fable 5.1 ↑2 18.0
  3. 3 GPT-6 Astra ↓1 18.0
  4. 4 Gemini 3.1 Pro ↓1 17.0
  5. 5 DeepSeek-V4-Pro ↑3 17.0
  6. 6 Gemini 3.8 Flash ↑1 17.0
  7. 7 GPT-6 Sol ↓2 17.0
  8. 8 Seed2.1 Pro ↑3 15.0
  9. 9 Qwen 3.8 Max ↓3 15.0
  10. 10 Tencent Hy3 ↓1 12.0

Number one over time

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

Anthropic
Own leaderboard 1 leader

Source: Official Hanabi Agent Benchmark 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
0.00 5.00 10.0 15.0 20.0 25.0 2026 0.00 5.00 10.0 15.0 20.0 25.0 2026

Source: Official Hanabi Agent Benchmark 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 Games and long tasks

All 261 in Games and long tasks
  • ARC-AGI-3 In use Interactive game environments where agents must explore, infer goals and rules, and act efficiently without instructions. GPT-6.2 Astra 100%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Each game scores in-game progress or outcomes, and the board ranks models by their average rank across the games. Gemini 2.5 Pro 3.90
  • MineDojo Fading Agents complete open-ended tasks in Minecraft, such as mining blocks and fighting mobs, from a large simulation suite. Programmatic checks score many tasks. Seed2.0 Pro 49%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. The score is harmful actions per 100 steps, where lower is better. The board also reports game progress. Llama 4 Scout 7.00
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. GPT-6.1 Terra 76.8%
  • Swarm Bench Only in system cards Models orchestrate agent swarms on complex tasks that benefit from coordinated decomposition and parallel execution. Kimi K3 76.3%

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 (21 entries).
  • Pass Models scored 19 models we track, plus 2 other entries on its leaderboard.
  • Pass Newest score The newest is from October 2026, for Claude Haiku 5.5.
  • Pass Frontier models 4 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on huggingface.co: 19 rows for models we track and 2 other entries 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