TokenAI Neo: 41M-parameter open-weights decision model
Alexandria-based TokenAI's sixth 2026 release: a tiny 'System One' decision model trained from scratch on 1M synthetic decisions that routes tool calls and escalates uncertain choices in one pass.
Why it matters
A 348MB decision model that outputs a choice, a score, and a yes/no in a single forward pass and flags low-confidence decisions for human review — cheap conservative routing for agent pipelines.
What you could build with it
An agent developer could drop Neo in as a cheap pre-filter that decides in a single pass which tool to call or whether to escalate to a human, instead of spending frontier-model tokens on every routing decision. A researcher could use its calibrated confidence outputs to study safe escalation policies for autonomous workflows.
Does it hold up?
Too early to judge: weights, training code, and dataset are public, but the custom non-commercial license keeps it in research territory and no third-party evaluations of its calibrated outputs exist yet.
Built with TokenAI Neo
- Middle East AI News: TokenAI launches compact AI decision modelblog · Analysis of Neo's 41M-param bidirectional-encoder design, its RLCD-inspired training, and the open question of how small a calibrated decision model can be.
Learn more
First spotted on article: source.
More AI models
Mistral Large 4 (Le Chonk): 1.05T-param open-weight MoEMistral's new flagship: a 1.05T-parameter mixture-of-experts multimodal model with 49B active params and 1M-token…models · JEV 0.72ElevenLabs Eleven v4 + v4 Turbo: new emotive TTS architecture with a 100ms real-time variantElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28: a new text-to-speech architecture with inline…models · JEV 0.72VoiceStudio: open-source, fully-local ElevenLabs alternative with voice cloning, dubbing and dictation in 646 languagesVoiceStudio is an open-source, fully-local voice AI studio: voice cloning from a 3-second sample, voice design, video…models · JEV 0.71Mistral Forge: enterprise platform for training and continuously improving proprietary modelsMistral AI launched Forge (Sep 28) — a full-lifecycle model training platform (pre-training, SFT, DPO/ODPO, RL…models · JEV 0.69DeepSeek V4-Flash official API: public beta with upgraded agent capabilities, Responses API and Codex supportDeepSeek's official V4-Flash is now a public beta API with massively upgraded agent capabilities - benchmark scores…models · JEV 0.69kev: open, trainable Jev-like family of small decision models on Qwen3.5/3.8A family of small decision models (0.8B to 27B) built on Qwen3.5/3.8 that reproduces Jev's prefill-only typed-decision…models · JEV 0.68
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.