Benchmark Almanac

Multi-Agent Arena Risk

GPT-6 Astra 38.8% leads since September 2026
Not in model cards

Four commanders, people or anonymized models, trade sealed messages and then attack for territory on a 42-territory map. Models are ranked by their mean share of territory, where 25 is an even split.

The board also rates how often a model’s attacks are gambles and how often it is two-faced in talks. It was scored from 106 games among models only, as of September 26th, 2026.

Made by
Olam Labs
Released
July 2026
Kind
Multi-agent games
Level
Everyday
Scored
Ranked by matches
Answers
Agent task
Tasks
Kept private
Measures
Capability
Language
English
In and out
Text → Actions, Text
Interaction
Multi-agent
Environment
Risk
Items from
Live play or votes
Refresh
Live
Run by
A host
In views
Games, Agents, Kept fresh
Strongest model tested
GPT-6 Astra #4 on the capability index
Last model added
September 2026
Scores refreshed
November 20th, 2026
16
Models scored on its own leaderboard

GPT-6 Astra leads with 38.8% 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-6 Astra 38.8%
  2. 2 GPT-6 Sol 34.3%
  3. 3 Claude Opus 5 33.5%
  4. 4 Gemini 3.8 Flash 33.5%
  5. 5 Claude Fable 5 33.3%
  6. 6 Claude Fable 5.1 32.2%
  7. 7 GPT-5.6 Sol 31%
  8. 8 Claude Opus 5.5 26.6%
  9. 9 Grok 4.7 23%
  10. 10 GLM-5.3 21.9%
  11. 11 Kimi K3 18.5%
  12. 12 GLM-5.3-Flash 18.4%
  13. 13 GPT-5.6 Terra 18.3%
  14. 14 Grok 4.6 15.6%
  15. 15 Muse Spark 1.3 14.4%
  16. 16 GPT-6 Luna 9.40%

Source: Official Multi-Agent Arena Risk leaderboard

#ModelScore ± 95% CISettingReleasedRun onSource
1 GPT-6 Astra 38.8% 29.7 to 48.8 Pi Sep 2026 Sep 2026 Official leaderboard
2 GPT-6 Sol 34.3% ±12.6 Pi Sep 2026 Sep 2026 Official leaderboard
3 Claude Opus 5 33.5% ±8 Pi Jul 2026 Sep 2026 Official leaderboard
4 Gemini 3.8 Flash 33.5% 25.2 to 42.2 Pi Sep 2026 Sep 2026 Official leaderboard
5 Claude Fable 5 33.3% ±8.9 Pi Jun 2026 Sep 2026 Official leaderboard
6 Claude Fable 5.1 32.2% 24 to 41.2 Pi Sep 2026 Sep 2026 Official leaderboard
7 GPT-5.6 Sol 31% ±9.2 Pi Jul 2026 Sep 2026 Official leaderboard
8 Claude Opus 5.5 26.6% 16.4 to 38.7 Pi Sep 2026 Sep 2026 Official leaderboard
9 Grok 4.7 23% ±4.6 Pi Sep 2026 Sep 2026 Official leaderboard
10 GLM-5.3 21.9% 14.9 to 29.2 Pi Aug 2026 Sep 2026 Official leaderboard
11 Kimi K3 18.5% 13.3 to 24.2 Pi Jul 2026 Sep 2026 Official leaderboard
12 GLM-5.3-Flash 18.4% 12.9 to 24.4 Pi Aug 2026 Sep 2026 Official leaderboard
13 GPT-5.6 Terra 18.3% 13.2 to 23.9 Pi Jul 2026 Sep 2026 Official leaderboard
14 Grok 4.6 15.6% 12.1 to 19.4 Pi Aug 2026 Sep 2026 Official leaderboard
15 Muse Spark 1.3 14.4% ±2.8 Pi Sep 2026 Sep 2026 Official leaderboard
16 GPT-6 Luna 9.40% 4.3 to 14.9 Pi Sep 2026 Sep 2026 Official leaderboard

3 models have held number one on the official leaderboard since its release in July 2026

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

Anthropic OpenAI
Own leaderboard Not tested since Sep 2026
GPT-6 Astra 38.8%

Source: Official Multi-Agent Arena Risk 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-6 Astra Claude Opus 5 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) GPT-6 Astra

Source: Official Multi-Agent Arena Risk 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 11 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 5 pass
  • Pass Independent runs Scores from the official leaderboard (16 entries).
  • Pass Models scored 16 models we track.
  • Pass Newest score The newest is from September 2026, for Claude Opus 5.5.
  • Pass Frontier models 3 of the top 10 have a score, led by GPT-6 Astra (#4).
  • Pass Headroom The best is 38.8%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on olamlabs.ai: 16 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