# Epoch AI reports rapid gains in AI play at Slay the Spire [Canonical reading](https://escapemeridian.com/readings/epoch-ai-reports-rapid-gains-in-ai-play-at-slay-the-spire-7b36bae7) On October 14, 2026, Epoch AI reports rapid gains in models’ play at Slay the Spire, while saying they remain far behind top human players at the hardest settings. Its SpireClimb benchmark tests models using a visual computer-use harness, meaning they play through the game interface. Epoch AI reports frontier models win at least 90% of attempts at the first game’s base difficulty, but GPT-6 Astra and GPT-6.1 Sol win 17% and 20% of attempts at its maximum difficulty. The sequel’s results are not directly equivalent: Epoch AI says its early-access version is easier and lacks a final boss comparable to The Heart. The publication interprets the sequel results as evidence that models do not take a significant penalty in a less familiar game, while noting that small character-level samples make firm conclusions difficult. SpireClimb also aims to measure what a person might get by asking a desktop AI app to play. Epoch AI argues that navigating a game interface while pursuing a goal may be relevant to automating human tasks, but the benchmark does not establish that models can do this outside games. Interface mistakes and the frozen sequel version limit what the tests capture. Headline and summary by Escape Meridian, prepared by Luna from the attributed source. A source check is not independent confirmation of the source’s claims. ## Original source Epoch AI · Article Original source date: Oct 14, 2026 (2026-10-14T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-31T11:32:16.203Z. First observed: 2026-10-18T02:34:57.095Z. Retrieved: 2026-10-23T18:13:40.918Z. - [AI systems are improving rapidly at Slay the Spire and its sequel](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Limits Text as retrieved on October 23, 2026; originally published October 14, 2026. The benchmark freezes the sequel at version v0.107.1 and does not track models against later versions played by the public. Epoch AI says interface mistakes occur and the visual computer-use harness may fail to elicit the full extent of models’ pure reasoning skills. Epoch AI notes that relatively low per-character sample sizes make firm conclusions difficult. The source does not report whether game-playing performance transfers to human tasks outside games. ## Open question How well will SpireClimb results track model performance as Slay the Spire 2 changes beyond the frozen version? ## Source-stated Epoch AI reports that frontier models win at least 90% of attempts at the first game's base difficulty. > Frontier models can now beat Slay the Spire 1 at base difficulty (“Ascension 0”) at least 90% of the time. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI reports GPT-6 Astra and GPT-6.1 Sol win rates of 17% and 20% at the first game's maximum difficulty. > with GPT-6 Astra achieving a 17% win rate and GPT-6.1 Sol achieving a 20% win rate. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI says top human players regularly achieve at least 80% win rates at maximum difficulty in both games. > These win rates remain far below those of top human players, who regularly achieve at least 80% win rates at maximum difficulty in both games. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI says the sequel's early-access version is broadly considered easier than the first game. > it is broadly considered much easier than the first game by most players. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI interprets the sequel results as evidence that frontier models do not take a significant penalty in a less familiar game. > Regardless, we interpret these results as evidence that frontier models do not take a significant penalty in playing a less familiar game. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI argues that skill in navigating game interfaces may be relevant to automating human tasks end-to-end, but presents this as a possibility, not a demonstrated result. > A model that can strategically navigate the interface of complex games while coherently pursuing its goals may also be better positioned to automate existing human tasks end-to-end - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source and contributor context Epoch AI · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## Frontier model One of the most capable AI models available at a given time; the set shifts as models improve. - [Definition reference](https://www.gov.uk/government/publications/frontier-ai-capabilities-and-risks-discussion-paper) ## Epoch AI Epoch AI studies trends in AI capability, training resources and the infrastructure behind them. - [Explore this context](/entities/epoch-ai) - [Reference](https://epoch.ai/) - [Reference](https://epoch.ai/data) ## Editorial provenance Version 1. Publication ID: 29630fb4-251d-d215-4fab-7aa279f2042a. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-31T11:32:12.877Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: 027b40e6031d07c52f9dd89b8bd39777492c59a66ec4b9301ecb40f60c3b2ca5. Source text hash: 82a243d99670ab28af0a907b0449494dbc25c611eb428b2a26e459b10b1b375e. ## Revision and source-check history ### Version 1 — 2026-10-31T11:32:16.203Z On October 14, 2026, Epoch AI reports rapid gains in models’ play at Slay the Spire, while saying they remain far behind top human players at the hardest settings. Its SpireClimb benchmark tests models using a visual computer-use harness, meaning they play through the game interface. Epoch AI reports frontier models win at least 90% of attempts at the first game’s base difficulty, but GPT-6 Astra and GPT-6.1 Sol win 17% and 20% of attempts at its maximum difficulty. The sequel’s results are not directly equivalent: Epoch AI says its early-access version is easier and lacks a final boss comparable to The Heart. The publication interprets the sequel results as evidence that models do not take a significant penalty in a less familiar game, while noting that small character-level samples make firm conclusions difficult. SpireClimb also aims to measure what a person might get by asking a desktop AI app to play. Epoch AI argues that navigating a game interface while pursuing a goal may be relevant to automating human tasks, but the benchmark does not establish that models can do this outside games. Interface mistakes and the frozen sequel version limit what the tests capture. ## Original source Epoch AI · Article Original source date: Oct 14, 2026 (2026-10-14T00:00:00.000Z). Date basis: Date supplied by the publisher’s feed. Added to Escape Meridian: 2026-10-31T11:32:16.203Z. First observed: 2026-10-18T02:34:57.095Z. Retrieved: 2026-10-23T18:13:40.918Z. - [AI systems are improving rapidly at Slay the Spire and its sequel](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Limits Text as retrieved on October 23, 2026; originally published October 14, 2026. The benchmark freezes the sequel at version v0.107.1 and does not track models against later versions played by the public. Epoch AI says interface mistakes occur and the visual computer-use harness may fail to elicit the full extent of models’ pure reasoning skills. Epoch AI notes that relatively low per-character sample sizes make firm conclusions difficult. The source does not report whether game-playing performance transfers to human tasks outside games. ## Open question How well will SpireClimb results track model performance as Slay the Spire 2 changes beyond the frozen version? ## Source-stated Epoch AI reports that frontier models win at least 90% of attempts at the first game's base difficulty. > Frontier models can now beat Slay the Spire 1 at base difficulty (“Ascension 0”) at least 90% of the time. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI reports GPT-6 Astra and GPT-6.1 Sol win rates of 17% and 20% at the first game's maximum difficulty. > with GPT-6 Astra achieving a 17% win rate and GPT-6.1 Sol achieving a 20% win rate. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI says top human players regularly achieve at least 80% win rates at maximum difficulty in both games. > These win rates remain far below those of top human players, who regularly achieve at least 80% win rates at maximum difficulty in both games. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI says the sequel's early-access version is broadly considered easier than the first game. > it is broadly considered much easier than the first game by most players. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI interprets the sequel results as evidence that frontier models do not take a significant penalty in a less familiar game. > Regardless, we interpret these results as evidence that frontier models do not take a significant penalty in playing a less familiar game. - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source-stated Epoch AI argues that skill in navigating game interfaces may be relevant to automating human tasks end-to-end, but presents this as a possibility, not a demonstrated result. > A model that can strategically navigate the interface of complex games while coherently pursuing its goals may also be better positioned to automate existing human tasks end-to-end - [Supporting source](https://epoch.ai/publications/ai-systems-are-improving-rapidly-at-slay-the-spire-and-its-sequel) ## Source and contributor context Epoch AI · Publisher The publisher is identified here. No individual author has been inferred. ## Source media Media retrieval status: unchecked. ## Terms and references Reference context selected by Escape Meridian; separate from source evidence. Entity mentions do not establish authorship, involvement or endorsement. ## Benchmark A defined set of tests used to compare systems. Results depend on the tasks and measures selected. - [Definition reference](https://arxiv.org/abs/2211.09110) ## Frontier model One of the most capable AI models available at a given time; the set shifts as models improve. - [Definition reference](https://www.gov.uk/government/publications/frontier-ai-capabilities-and-risks-discussion-paper) ## Epoch AI Epoch AI studies trends in AI capability, training resources and the infrastructure behind them. - [Explore this context](/entities/epoch-ai) - [Reference](https://epoch.ai/) - [Reference](https://epoch.ai/data) ## Editorial provenance Version 1. Publication ID: 29630fb4-251d-d215-4fab-7aa279f2042a. Prepared by openai/gpt-6-luna; grounding review by anthropic/claude-opus-5.5. Source check: passed by anthropic/claude-opus-5.5 on 2026-10-31T11:32:12.877Z. This checks fidelity to the source, not whether its claims are independently true. Editorial selection: accept; reasons: bridge\_admitted\_by\_model. Observation: 027b40e6031d07c52f9dd89b8bd39777492c59a66ec4b9301ecb40f60c3b2ca5. Source text hash: 82a243d99670ab28af0a907b0449494dbc25c611eb428b2a26e459b10b1b375e. ### passed — 2026-10-31T11:32:12.877Z — anthropic/claude-opus-5.5 Publication: 29630fb4-251d-d215-4fab-7aa279f2042a attribution — pass: This is a single Epoch AI publication presenting its SpireClimb benchmark. The reading attributes findings, interpretations and the automation argument to Epoch AI throughout. It does not present them as established fact. > Here, we present findings from our new benchmark, SpireClimb, that show this is rapidly changing. summary — pass: The reading leads with the actor (Epoch AI) and the development (rapid gains, still far below top humans). It explains the visual computer-use harness in plain terms, and benchmark and frontier model are defined beside the reading. It keeps Epoch's force on rapid improvement while preserving its stated limits. The question about the frozen sequel version follows from the source's own acknowledged drift. The sequel's above-25% figure is omitted, but its non-equivalence is explained, and that omission does not distort the meaning. > This means that over time, the version of StS2 we use for this benchmark will drift away from the version the public actually plays. date — pass: The reading states October 14, 2026, which matches the UTC calendar day of evidence.originalPublished. The evidence is a publisher page, so no archive disclosure is needed. Future plans, such as evaluating new models and possible saturation, are not presented as outcomes. The reading uses no hindsight framing or later events. support — pass: Each claim is supported by a verbatim excerpt: the 90% base-difficulty rate, the 17% and 20% maximum-difficulty rates, the at-least-80% human rates, the sequel being easier, and the less-familiar-game interpretation. The automation link is kept conditional, as the source frames it ('may also be better positioned'). The reading's caveat that this is not demonstrated outside games does not overstate the source. The limitations on interface mistakes, sample sizes and the frozen version are all sourced. > Interface mistakes still occur, and requiring models to play the game through its graphical interface may encourage quicker decisions at the expense of more deliberate reasoning. media — pass: No media was selected, and no verified media candidates were supplied. A null selection is appropriate. ## Recorded references Explicit citations in public readings and public file attachments to this reading. Shared topics are not citations. Other artifact types need an explicit reference adapter. No public references recorded here yet. [Reference data (JSON)](/readings/7b36bae79eea7aadb353a73bc17e7b92650b49ddf37b50991ef4f9fec6fa70df/references.json) ## Corrections and challenges [Send a correction or an opposing source](https://escapemeridian.com/corrections/) [JSON record](https://escapemeridian.com/readings/epoch-ai-reports-rapid-gains-in-ai-play-at-slay-the-spire-7b36bae7.json)