GPT-6-Astra
Timeline (last 90 days)
Deep Principle claims Hydrogen-1.0 outperforms GPT-6-Astra on the FrontierScience-Research benchmark, but the result lacks third-party reproduction.
The research team tested 12 LLMs, and only GPT-6-Astra achieved a discovery rate of 16.7%, while the remaining 11 models all had a discovery rate of 0%.
Against GPT-oss-120B, GPT-5.4, Claude-Haiku-4.5 and GPT-6-Astra, the adversarially trained SEMA-Auto achieves an average of 59.6% ASR@1 on HarmBench.
In direct control mode, GPT-6-Astra's overall performance is competitive with post-trained policies, leads on short-horizon tasks, and shows the smallest performance fluctuation across generalization dimensions.
The EBench evaluation report centers on GPT-6-Astra, building an agent that controls a robot through an agent harness and comparing it using the exact same visual and proprioceptive observation protocol as the post-trained policies.
In direct control mode, GPT-6-Astra lags behind post-trained policies on long-horizon tasks, tabletop manipulation, and high-precision contact-intensive tasks.
When the agent is allowed to invoke post-trained VLA or WAM policies, overall task success rate rises to the level of the strongest comparison policy, while token cost drops significantly.
Even the frontier model GPT-6-Astra can solve only 60% of the WildAWM tasks.
The participating models included Opus 5.5 and Fable 5.1 accessed via Claude Code, GPT-6-Astra and GPT-6.1-Sol accessed via Codex, and Gemini 4 Argon (High) and Gemini 3.8 Flash (High) accessed via Gemini CLI.
GPT-6-Astra scored 2 points but had 2 strikes; Fable 5.1, Gemini 3.8 Flash and GPT-6.1-Sol each scored 1 point and all accumulated 3 strikes.
1 more timeline events
All Facts (7)
Deep Principle claims Hydrogen-1.0 outperforms GPT-6-Astra on the FrontierScience-Research benchmark, but the result lacks third-party reproduction.
50%UnverifiedThe research team tested 12 LLMs, and only GPT-6-Astra achieved a discovery rate of 16.7%, while the remaining 11 models all had a discovery rate of 0%.
50%UnverifiedAgainst GPT-oss-120B, GPT-5.4, Claude-Haiku-4.5 and GPT-6-Astra, the adversarially trained SEMA-Auto achieves an average of 59.6% ASR@1 on HarmBench.
50%UnverifiedThe EBench evaluation report centers on GPT-6-Astra, building an agent that controls a robot through an agent harness and comparing it using the exact same visual and proprioceptive observation protocol as the post-trained policies.
50%UnverifiedWhen the agent is allowed to invoke post-trained VLA or WAM policies, overall task success rate rises to the level of the strongest comparison policy, while token cost drops significantly.
50%UnverifiedIn direct control mode, GPT-6-Astra's overall performance is competitive with post-trained policies, leads on short-horizon tasks, and shows the smallest performance fluctuation across generalization dimensions.
50%UnverifiedIn direct control mode, GPT-6-Astra lags behind post-trained policies on long-horizon tasks, tabletop manipulation, and high-precision contact-intensive tasks.
50%