Open-source Tess-4.5-594B beats base model on SWE-Bench Pro Hard with 10% fewer tokens
Scobleizer · x · 2026-11-14
migtissera released Tess-4.5-594B-A17B, an open-source model post-trained from Thinking Machines' Inkling 2 Small using the Tess-4.5 recipe on real engineering sessions, targeting dense reasoning, an empathetic tone, and hardcore engineering. On SWE-Bench Pro v2 Hard it resolves 88.2% of the hardest public software-engineering tasks vs 86.3% for the base model, while using 10% fewer tokens. The weights are on Hugging Face, with observers noting open source's continued momentum.
More from Models
- 4B specialized model hits 93.8% on bank support, beating proprietary models' 74% — MParakhin · 2026-11-14
- AI Builder Pushes Back on AGI-2027: Jaggedness Persists, Models Still Ignore Simple Instructions — ChrisGPT · 2026-11-14
- Claude models decide by emotivism but rationalize post-hoc, researcher observes — Sauers_ · 2026-11-14
- Heavy user's first impressions: Fable 5.2 is the best model, but Opus 5.5 makes more sense for heavy token users — jarrodwatts · 2026-11-14
- Open ASR model hits #10 overall on Hugging Face leaderboard, 2nd among open models — realmrfakename · 2026-11-14
- Fable 5.2 built a playable 3D PC-building simulator from scratch, all models AI-generated — OneKey3719 · 2026-11-14