Benchmark Almanac

PTCG-Bench

Gemini 3.1 Pro 1,854 leads since May 2026
Not in model cards

Agents play the Pokémon Trading Card Game by choosing legal actions from a partially visible board. Match results determine their Glicko-2 ratings.

The authors rank ten language models and report costs per agent. A fixed strategy and a random player occupy two additional rows.

Released
May 2026
Kind
Multi-agent games
Level
School
Scored
Ranked by matches
Answers
Agent task
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Actions
Interaction
Multi-agent
Items from
Live play or votes
Refresh
Live
Run by
Anyone
In views
Games, Agents, Kept fresh
Strongest model tested
GPT-5.4 #28 on the capability index
Last model added
April 2026
Scores refreshed
November 20th, 2026
10
Models scored on its own leaderboard

Gemini 3.1 Pro leads with 1,854 among models released by April 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. Every run is in the next chart.

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

Bars start at 1,200.

  1. 1 Gemini 3.1 Pro 1,854
  2. 2 DeepSeek-V4-Flash 1,727
  3. 3 DeepSeek-V4-Pro 1,727
  4. 4 Gemini 3 Flash 1,636
  5. 5 Claude Sonnet 4.6 1,600
  6. 6 GPT-5.4 1,564
  7. 7 Qwen 3.6 Plus 1,509
  8. 8 Claude Haiku 4.5 1,473
  9. 9 Qwen3.5-Flash 1,382
  10. 10 GPT-5.4 Nano 1,237

Source: Official PTCG-Bench leaderboard

#ModelScoreSettingDateSource
1 Gemini 3.1 Pro 1,854 Feb 2026 Official leaderboard
2 DeepSeek-V4-Flash 1,727 Apr 2026 Official leaderboard
3 DeepSeek-V4-Pro 1,727 Apr 2026 Official leaderboard
4 Gemini 3 Flash 1,636 Dec 2025 Official leaderboard
5 Claude Sonnet 4.6 1,600 Feb 2026 Official leaderboard
6 GPT-5.4 1,564 Mar 2026 Official leaderboard
7 Qwen 3.6 Plus 1,509 Mar 2026 Official leaderboard
8 Claude Haiku 4.5 1,473 Oct 2025 Official leaderboard
9 Qwen3.5-Flash 1,382 Feb 2026 Official leaderboard
10 GPT-5.4 Nano 1,237 Mar 2026 Official leaderboard

The best score, 1,854, costs $1.15 a task

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.

Google DeepMind Qwen OpenAI DeepSeek Cost frontier One model at each effort level
0 500 1,000 1,500 2,000 $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 3.1 Pro DeepSeek-V4-Flash Qwen3.5-Flash GPT-5.4 Nano 0 500 1,000 1,500 2,000 $0.01 $0.10 $1.00 $10 Cost per task, log scale Gemini 3.1 Pro DeepSeek-V4-Flash

Source: Official PTCG-Bench leaderboard

ModelEffortHarnessScoreCost per taskOutput tokens
Gemini 3.1 Pro 1,854 $1.15
DeepSeek-V4-Flash 1,727 $0.09
DeepSeek-V4-Pro 1,727 $0.23
Gemini 3 Flash 1,636 $0.25
Claude Sonnet 4.6 1,600 $1.94
GPT-5.4 1,564 $0.68
Qwen 3.6 Plus 1,509 $0.17
Claude Haiku 4.5 1,473 $0.90
Qwen3.5-Flash 1,382 $0.05
GPT-5.4 Nano 1,237 $0.04

Number one over time

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

Google DeepMind
Own leaderboard Not tested since May 2026

Source: Official PTCG-Bench 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
1,200 1,400 1,600 1,800 2,000 2026 1,200 1,400 1,600 1,800 2,000 2026

Source: Official PTCG-Bench 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

8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (10 entries).
  • Pass Models scored 10 models we track.
  • Warning Newest score The newest came 7 months ago, in April 2026, for DeepSeek-V4-Flash.
  • Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5.4 (#28).
  • Note Headroom Scored as a rating, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on ddhua.tech: 10 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