Benchmark Almanac

CUA-bench

GPT-6 Astra 19.2% leads since September 2026
Not in model cards

Computer-use agents play six commercial video games, such as Minecraft, SUPERHOT and eFootball, using only a keyboard and mouse. Half of the games are held out.

Vals reports estimated API cost and time per action. The board had nine models on October 13th, 2026.

Made by
Vals AI
Released
September 2026
Kind
Video games
Level
Everyday
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
GPT-6 Astra #3 on the capability index
Last model added
October 2026
Scores refreshed
November 12th, 2026
9
Models scored on its own leaderboard

GPT-6 Astra leads with 19.2%

Best score per model from the benchmark's own leaderboard, read on November 12th, 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart.

  1. 19.2 GPT-6 Astra
  2. 14.0 Claude Opus 5.5
  3. 13.2 Claude Fable 5.1
  4. 9.00 Claude Opus 5
  5. 8.33 GPT-5.6 Sol
  6. 7.17 Muse Wonder
  7. 5.83 Muse Spark 1.3
  8. 4.83 Gemini 4 Argon
  9. 4.17 Gemini 3.8 Flash
  1. 1 GPT-6 Astra 19.2%
  2. 2 Claude Opus 5.5 14%
  3. 3 Claude Fable 5.1 13.2%
  4. 4 Claude Opus 5 9.00%
  5. 5 GPT-5.6 Sol 8.33%
  6. 6 Muse Wonder 7.17%
  7. 7 Muse Spark 1.3 max 5.83%
  8. 8 Gemini 4 Argon 4.83%
  9. 9 Gemini 3.8 Flash 4.17%

Source: Official CUA-bench leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 19.2% Sep 2026 Official leaderboard
2 Claude Opus 5.5 14% Sep 2026 Official leaderboard
3 Claude Fable 5.1 13.2% Sep 2026 Official leaderboard
4 Claude Opus 5 9.00% Jul 2026 Official leaderboard
5 GPT-5.6 Sol 8.33% Jul 2026 Official leaderboard
6 Muse Wonder 7.17% Oct 2026 Official leaderboard
7 Muse Spark 1.3 5.83% max Sep 2026 Official leaderboard
8 Gemini 4 Argon 4.83% Sep 2026 Official leaderboard
9 Gemini 3.8 Flash 4.17% Sep 2026 Official leaderboard

The best score, 19.2%, costs $1,821 for the full run

Runs on the cost frontier score higher than any run that's cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

Meta OpenAI Anthropic Cost frontier One model at each effort level
0% 5% 10% 15% 20% $100 $200 $500 $1,000 $2,000 $5,000 $10,000 Cost of the full run, log scale GPT-6 Astra Claude Opus 5.5 Muse Spark 1.3, max 0% 5% 10% 15% 20% $100 $1,000 $10,000 Cost of the full run, log scale GPT-6 Astra Claude Opus 5.5

Source: Official CUA-bench leaderboard

ModelEffortHarnessScoreCost of the full runOutput tokens
GPT-6 Astra 19.2% $1,821
Claude Opus 5.5 14% $188
Claude Fable 5.1 13.2% $324
Claude Opus 5 9.00% $232
GPT-5.6 Sol 8.33% $207
Muse Wonder 7.17% $597
Muse Spark 1.3 max 5.83% $144
Gemini 4 Argon 4.83% $194
Gemini 3.8 Flash 4.17% $280

2 models have held number one on the official leaderboard since its release in September 2026

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

Anthropic OpenAI
Own leaderboard 2 leaders
GPT-6 Astra 19.2%

Source: Official CUA-bench leaderboard

Scores over time

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

On the benchmark’s own leaderboard Best so far
0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1 Claude Opus 5 0% 20% 40% 60% 80% 100% Near the ceiling (85%) GPT-6 Astra Claude Opus 5

Source: Official CUA-bench leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

More in Games and long tasks

All 82 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 Astra 99.95%
  • 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%