BenchLeader
Menu
LeaderboardsModelsProvidersBenchmarksCompareReleases1BlogMethodology

Poker Agent

AI agents playing poker against each other. Run by Vals AI.

As of 19 Sept 2026, GPT-5.2 leads Poker Agent on BenchLeader with 1131.8%, ahead of GPT-5 at 1103.2%, across 16 model configurations with a published result.

Published by
Vals AI
Category
Agents & tools
Index weight
Reference only
Models
16
Data as of
19 Sept 2026

Vals AI (vals.ai).

What the test looks like

Models play many hands of poker against one another and are scored on winnings, testing reasoning under uncertainty and opponent modelling.

How it is scored

Relative winnings across matches, run by Vals AI.

What to keep in mind

A tournament score depends on the field; compare within one snapshot.

Top 15
  1. 1GPT-5.21131.8%
  2. 2GPT-51103.2%
  3. 3Gemini 3 Flash1100.2%
  4. 4DeepSeek V3.2 (thinking)1090.3%
  5. 5Grok 4.1 (thinking)1079.2%
  6. 6Gemini 3 Pro1078.9%
  7. 7Claude Sonnet 4.51055.5%
  8. 8GPT-5.11038.6%
  9. 9Grok 4 Fast (thinking)1034.3%
  10. 10Claude Opus 4.51033.4%
  11. 11Gemini 2.5 Pro1032.6%
  12. 12gpt-oss-120b1015.3%
  13. 13Kimi K2 Thinking (thinking)1011.6%
  14. 14Qwen3 Max (max)994.5%
  15. 15GLM-4.6945.8%
16 of 16
#
1GPT-5.2OpenAI1131.8%58.42025-12-23
2GPT-5OpenAI1103.2%57.22025-12-23
3Gemini 3 FlashGoogle1100.2%57.72025-12-23
4DeepSeek V3.2thinkingDeepSeekopen1090.3%56.12025-12-23
5Grok 4.1thinkingxAI1079.2%56.02025-12-23
6Gemini 3 ProGoogle1078.9%61.12025-12-23
7Claude Sonnet 4.5Anthropic1055.5%54.32025-12-23
8GPT-5.1OpenAI1038.6%57.22025-12-23
9Grok 4 FastthinkingxAI1034.3%52.62025-12-23
10Claude Opus 4.5Anthropic1033.4%58.22025-12-23
11Gemini 2.5 ProGoogle1032.6%54.52025-12-23
12gpt-oss-120bOpenAIopen1015.3%46.92025-12-23
13Kimi K2 ThinkingthinkingMoonshot AIopen1011.6%52.22025-12-23
14Qwen3 MaxmaxAlibaba994.5%53.52025-12-23
15GLM-4.6Zhipu AIopen945.8%49.82025-12-23
16Llama 4 Maverick BasicMeta890.5%39.42025-12-23

Cite as: BenchLeader, “Poker Agent leaderboard”, https://www.benchleader.com/benchmarks/vals_poker, data as of 19 Sept 2026.

Poker Agent: questions

What does Poker Agent measure?
Models play many hands of poker against one another and are scored on winnings, testing reasoning under uncertainty and opponent modelling. Scores are reported in percent of tasks solved; higher is better.
Which AI model leads Poker Agent?
GPT-5.2 leads Poker Agent with 1131.8% as of 19 Sept 2026, ahead of GPT-5 at 1103.2%.
How many models have Poker Agent results?
16 model configurations have a Poker Agent result on BenchLeader, all taken from Vals AI.
Who runs Poker Agent and how often is it updated?
Poker Agent is published by Vals AI. BenchLeader re-reads the published results every morning and records the date each result was published.
Does Poker Agent count toward the BenchLeader Index?
No. Poker Agent is shown for reference but left out of the composite index.