Epoch AI saturated the EBR-bench long-horizon gaming benchmark with two playthroughs, scoring 100%, outpacing the top human player in the baseline.
Public source
Publisher name
Public post
GPT-6 Astra has set a new ECI record, with a score of 169. This is a substantial jump from the prior best (163), but is within our uncertainty range for the reasoning-er…
Company
Epoch AI
Research institute investigating the trajectory of AI
- Industry
- Research Services
- Location
- San Francisco, US
- Company size
- 11–50 employees
About Epoch AI
Epoch AI is a multidisciplinary research institute investigating the trajectory of Artificial Intelligence (AI). We scrutinize the driving forces behind AI and forecast its ramifications on the economy and society. We emphasize making our research accessible through our reports, models and visualizations to help ground the discussion of AI on a solid empirical footing. Our goal is to create a healthy scientific environment, where claims about AI are discussed with the rigor they merit.
See moreLatest activity
Latest activity from Epoch AI
9 signals
Research & Knowledge
Epoch AI notes that the sample size is small, and new plans may materialize or existing ones may be expanded.
Research & Knowledge
Epoch AI estimates that the IT power capacity of the largest data center has been growing by approximately 2.3x per year through early 2027.
Research & Knowledge
Epoch AI estimated Huawei could ship about 1.5 million Ascend chips in 2026 using LPDDR memory.
Discover more
Similar signals
Similar public activity from other companies.
Research & Knowledge
Google DeepMind
Google DeepMind's Gemini 3.7 Flash reached Victory Road in Pokémon in just ~22,000 turns compared to ~70,000 turns for 3.5 Flash.
Research & Knowledge
Apodex
Apodex 1.1 (Agent Team) achieved a 63.3% pass rate across 60 tasks on FrontierScience-Research with a rubric score of ≥7/10 counted as a pass.
Research & Knowledge
LaunchReady.ai
LaunchReady.ai reported that Agentic Scientific Research reached 52.6% benchmarks on Terminal-Bench-Science 0.1.
Research & Knowledge
ulam.ai
ulam.ai's GPT-6 Astra xhigh scored first on the ErdosBench benchmark of 226 open math problems inspired by Erdos problems.
Research & Knowledge
OpenAI