Benchmark Almanac

BalatroBench

Gemini 3 Pro 19.3 leads since Nov 2025 Not in model cards

Models play the deck-building roguelike Balatro through an API bot. The leaderboard compares how far each model and strategy gets in a run.

Built on Coder's BalatroBot API and BalatroLLM runner. Run data is on Kaggle.

Made by Coder Released Aug 2025 Video games All tasks public Last model added Jan 2026 Scores refreshed 2 Nov 2026
Level
Everyday Most adults, without training
Scored
Checked automatically A program checks the result against an answer key, unit tests or the expected final state.
Answers
Agent task Many steps in a terminal, browser, desktop, app, game or robot.
coder/balatrobench 23 stars Dataset Leaderboard balatrobench.com
17
Entries on its own leaderboard, 16 models we track

Gemini 3 Pro leads with 19.3 among models released by Jan 2026

Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.

The newest model with a score here came out in Jan 2026, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

17 entries. Scroll sideways for the rest.

  1. 19.3 Gemini 3 Pro
  2. 18.4 GPT-5.2
  3. 16.7 Gemini 3 Flash
  4. 13.9 Claude Opus 4.5
  5. 11.9 Claude Sonnet 4.5
  6. 9.47 Grok 4
  7. 8.67 DeepSeek-V3.2
  8. 7.67 Grok 4.1 Fast
  9. 7.27 Qwen3-Max
  10. 6.27 gpt-oss-120b
  11. 5.80 gpt-oss-20b
  12. 5.47 Kimi K2 Thinking
  13. 5.33 Kimi K2.5
  14. 5.07 Claude Haiku 4.5
  15. 3.07 GLM-4.7
  16. 3.07 MiniMax-M2.1
  17. 1.47 mistral-large-2512
  1. 1 Gemini 3 Pro 19.3
  2. 2 GPT-5.2 18.4
  3. 3 Gemini 3 Flash 16.7
  4. 4 Claude Opus 4.5 13.9
  5. 5 Claude Sonnet 4.5 11.9
  6. 6 Grok 4 9.47
  7. 7 DeepSeek-V3.2 8.67
  8. 8 Grok 4.1 Fast 7.67
  9. 9 Qwen3-Max 7.27
  10. 10 gpt-oss-120b 6.27
  11. 11 gpt-oss-20b 5.80
  12. 12 Kimi K2 Thinking 5.47
  13. 13 Kimi K2.5 5.33
  14. 14 Claude Haiku 4.5 5.07
  15. 15 GLM-4.7 3.07
  16. 16 MiniMax-M2.1 3.07
  17. 17 mistral-large-2512 1.47
#EntryScoreSettingDateSource
1 Gemini 3 Pro 19.3 Nov 2025 Official leaderboard
2 GPT-5.2 18.4 Dec 2025 Official leaderboard
3 Gemini 3 Flash 16.7 Dec 2025 Official leaderboard
4 Claude Opus 4.5 13.9 Nov 2025 Official leaderboard
5 Claude Sonnet 4.5 11.9 Sep 2025 Official leaderboard
6 Grok 4 9.47 Jul 2025 Official leaderboard
7 DeepSeek-V3.2 8.67 Dec 2025 Official leaderboard
8 Grok 4.1 Fast 7.67 Nov 2025 Official leaderboard
9 Qwen3-Max 7.27 Sep 2025 Official leaderboard
10 gpt-oss-120b 6.27 Aug 2025 Official leaderboard
11 gpt-oss-20b 5.80 Aug 2025 Official leaderboard
12 Kimi K2 Thinking 5.47 Nov 2025 Official leaderboard
13 Kimi K2.5 5.33 Jan 2026 Official leaderboard
14 Claude Haiku 4.5 5.07 Oct 2025 Official leaderboard
15 GLM-4.7 3.07 Dec 2025 Official leaderboard
16 MiniMax-M2.1 3.07 Dec 2025 Official leaderboard
17 mistral-large-2512 1.47 Official leaderboard

The best score, 19.3, costs $11 a task

Runs on the cost frontier score higher than any run that's cheaper. GPT-5.2 comes within 0.9 of the best for $4.36 a task, 3 times cheaper. The costs are the leaderboard's own estimates, so compare them only within this benchmark.

OpenAI Google DeepMind xAI DeepSeek Cost frontier One model at each effort level
0.00 5.00 10.0 15.0 20.0 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 3 Pro GPT-5.2 Gemini 3 Flash DeepSeek-V3.2 Grok 4.1 Fast gpt-oss-20b 0.00 5.00 10.0 15.0 20.0 $0.01 $0.10 $1.00 $10 $100 Cost per task, log scale Gemini 3 Pro Gemini 3 Flash
ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 3 Pro 19.3 $11
GPT-5.2 18.4 $4.36
Gemini 3 Flash 16.7 $2.15
Claude Opus 4.5 13.9 $15
Claude Sonnet 4.5 11.9 $8.88
Grok 4 9.47 $6.27
DeepSeek-V3.2 8.67 $1.01
Grok 4.1 Fast 7.67 $0.22
Qwen3-Max 7.27 $1.14
gpt-oss-120b 6.27 $0.27
gpt-oss-20b 5.80 $0.09
Kimi K2 Thinking 5.47 $0.64
Kimi K2.5 5.33 $0.26
Claude Haiku 4.5 5.07 $1.19
GLM-4.7 3.07 $0.19
MiniMax-M2.1 3.07 $0.11

3 models have held number one since its release in Aug 2025

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

xAI Anthropic Google DeepMind
Own leaderboard Not tested since Jan 2026
Gemini 3 Pro 19.3

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
0.00 5.00 10.0 15.0 20.0 2026 0.00 5.00 10.0 15.0 20.0 2026

No lab model card reports it yet

We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.

More in Games and long tasks

All 68 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 100%
  • 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%
  • Orak Fading LLM agents play 12 popular video games across most genres through an MCP interface. Scores come from in-game progress or outcomes per title. Nemotron 3 Nano 4B 22.9
  • GameWorld Only in system cards Multimodal game agents play 34 browser games across 170 tasks. Outcomes are checked from the game state. Gemini 3 Flash 41.9%
  • 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%
  • TextQuests Only in system cards Agents play 25 classic interactive fiction games. Measures game progress and harmful actions per 100 steps. Muse Spark 2.40