A foundation model that outputs probabilities, not words: 25x faster, ~600x cheaper
danshipper · x · 2026-09-16
Dan Shipper shares that the team at Every has been testing a new kind of foundation model by typesafeai: instead of producing words, it outputs probabilities — like a code linter for knowledge work. In their testing it ran about 25x faster at nearly 600x lower cost than Fable for similar jobs. Use cases include judging whether code meets your standards, whether writing contains AI-isms, or whether you'd be interested in a given tweet. Shipper notes they rarely test new flavors of foundation models that end up impressive — this is one of them.
Related event: TypeSafe's Probability-Output Model Runs 25x Faster at 600x Lower Cost(4 posts)→
More from coding & agent
- Combining Continued Pretraining with RAG: Teaching Qwen 3.5 4B a Fictional Subway Map — funJS · 2026-09-16
- ChatGPT $100-Tier User: GPT-6 Astra Burns 3x Usage Budget Yet Codes Worse Than GPT-5.6 Sol — F0xy1337 · 2026-09-16
- Plane to launch workspace Agents this week — JosephJacks_ · 2026-09-16
- GlossoGen: open-source platform shows AI agents spontaneously inventing new languages under pressure — EliasEskin · 2026-09-16
- Attach Claude, Codex and Gemini agents to RealSense cameras with one npx command — chrismatthieu · 2026-09-16
- One memory config across multiple agent types will quietly ruin all of them — Adventurous_Whole973 · 2026-09-16