Open-source TwIL-8B: 8.8B params hits 90%+ of frontier scores, runs locally
dair_ai · x · 2026-11-05
TwIL-8B is now open source: just 8.8B parameters yet approaching frontier-model scores — 93% of frontier performance on coding, 92% on instruction following, 90% on contest math — small enough to run on your own hardware.
The author argues the industry's focus is shifting from raw intelligence to the cost of completing work, and that efficient small models will power complex agentic workflows like scientific discovery and software factories. dair-ai's Omar Sansevero shared the release, bullish on locally runnable small models.
More from Models
- Kimi K3.1 Pro hits Engy API at $1.40/$7.00 per 1M tokens, 30% off — const_reborn · 2026-11-05
- Ripple-65B open model tops Terminal-Bench 2.1 at 80.5%, trained from scratch on $32M compute — annadgoldie · 2026-11-05
- Datalab tested 15 document-parsing benchmarks: about half misrank parsers vs real errors — VikParuchuri · 2026-11-05
- SkyPilot Tokens lists Kimi K3.1 Pro day-0: 2.8T params, 1M context, 40% cheaper — skypilot_org · 2026-11-05
- New open-source American LLM pretrained from scratch teased, drawing AI community buzz — annadgoldie · 2026-11-05
- OpenAI October dev roundup: Decisions API, 12-hour caching, 8x faster Codex tokens — OpenAIDevs · 2026-11-05