According to StanfordAILab, OpenAI’s GPT6 Astra scores 64.6% on Terminal Bench Science, up 42.2pp from GPT5.6 Sol, per StevenDillmann’s announcement.

According to StanfordAILab, Terminal-Bench-Science v0.1 tests 70 research tasks and Claude Opus 5 solves about 30%, signaling big gaps in agent tooling.

According to JeffDean, Gemini 3.5 Flash beats 3.1 Pro on agentic and coding benchmarks and runs 4x faster than frontier models, enabling scalable sub-agents.