Benchmark Almanac

SpireClimb StS1 A20 Heart

GPT-6 Astra 33.3% leads since October 2026
Not in model cards

Agents play Slay the Spire 1 through its graphical interface at Ascension 20 and must defeat the Heart. Scored by win rate over fixed seeds.

Epoch expects saturation within a few model generations because labs may train on the game.

Made by
Epoch AI
Released
October 2026
Kind
Video games
Level
Expert
Scored
Checked automatically
Answers
Agent task
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last model added
November 2026
Scores refreshed
November 12th, 2026
9
Models scored on its own leaderboard

GPT-6 Astra leads with 33.3%

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.

  1. 33.3 GPT-6 Astra
  2. 21.7 Claude Fable 5.2
  3. 18.3 GPT-6.1 Sol
  4. 10.0 GPT-6.1 Terra
  5. 5.00 Claude Opus 5.5
  6. 5.00 GPT-6 Sol
  7. 0.0 Claude Fable 5.1
  8. 0.0 GPT-6 Luna
  9. 0.0 Claude Sonnet 5.5
  1. 1 GPT-6 Astra max; task version 0.3.3_cd56aef 33.3%
  2. 2 Claude Fable 5.2 max; task version 0.3.3_cd56aef 21.7%
  3. 3 GPT-6.1 Sol max; task version 0.3.3_cd56aef 18.3%
  4. 4 GPT-6.1 Terra max; task version 0.3.3_cd56aef 10%
  5. 5 Claude Opus 5.5 max; task version 0.3.3_cd56aef 5.00%
  6. 6 GPT-6 Sol max; task version 0.3.3_cd56aef 5.00%
  7. 7 Claude Fable 5.1 max; task version 0.3.3_cd56aef 0%
  8. 8 GPT-6 Luna max; task version 0.3.3_cd56aef 0%
  9. 9 Claude Sonnet 5.5 max; task version 0.3.3_cd56aef 0%

Source: Official SpireClimb StS1 A20 Heart leaderboard

#ModelScoreSettingDateSource
1 GPT-6 Astra 33.3% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
2 Claude Fable 5.2 21.7% max; task version 0.3.3_cd56aef, Epoch Nov 2026 Official leaderboard
3 GPT-6.1 Sol 18.3% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
4 GPT-6.1 Terra 10% max; task version 0.3.3_cd56aef, Epoch Oct 2026 Official leaderboard
5 Claude Opus 5.5 5.00% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
6 GPT-6 Sol 5.00% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
7 Claude Fable 5.1 0% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
8 GPT-6 Luna 0% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard
9 Claude Sonnet 5.5 0% max; task version 0.3.3_cd56aef, Epoch Sep 2026 Official leaderboard

Epoch AI has run 9 models. We plot every source in the charts below.

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release. We draw one row each for the benchmark’s own leaderboard and Epoch AI’s runs.

OpenAI
Own leaderboard 1 leader
GPT-6 Astra 33.3%
Epoch AI runs 1 leader
GPT-6 Astra 33.3%

Sources: Official SpireClimb StS1 A20 Heart leaderboard and Epoch AI runs (CC BY 4.0)

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 Run by Epoch AI Best so far
0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1 0% 20% 40% 60% 80% 100% 2027 Near the ceiling (85%) GPT-6 Astra Claude Fable 5.1

Sources: Official SpireClimb StS1 A20 Heart leaderboard and Epoch AI runs (CC BY 4.0)

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%