New model 'Astra' claims massive improvement over GPT-5.6 on ExploitBench
daniel_mac8 · x · 2026-09-02
Developer announced 'Astra' is ready, claiming an "ungodly improvement" over GPT-5.6 Sol on an internal version of ExploitBench. The tweet suggests significant breakthroughs in specific tasks, likely related to security or vulnerability exploitation.
More from Models
- Grok 4.6 retains #1 spot on GPQA Diamond leaderboard — XFreeze · 2026-09-02
- Amp ultra mode upgrades to Fable 5.1 with better performance and 35% lower cost — HankYeomans · 2026-09-02
- Fable 5.1 hits 33k lines on delete code bench — Sauers_ · 2026-09-02
- View: Abliteration less expressive than training — maximelabonne · 2026-09-02
- Fable 5.1 tops Bug Hunt Bench, beating GPT-5.6 with better speed and cost — PawelHuryn · 2026-09-02
- The Hugging Face incident isn't isolated: supply-chain worries over open-source models — StewartalsopIII · 2026-09-02