Pareto 26.10 Preview: composite model from Unbiased (Circuit and Chisel)
Preview release of Pareto, a blended AI model that runs several frontier and open models in parallel on every request and synthesizes one answer.
Why it matters
A composite model rather than a router: every request engages multiple models and their results are combined per-task, so long agentic trajectories keep prompt-caching savings that routers lose when they switch models mid-conversation. Unbiased's preliminary benchmarks put it level with Claude Fable 5 on DeepSWE v1.1 at $0.24 per task vs $13.50.
What you could build with it
A startup building a developer support agent could route its long debugging sessions through a composite model like Pareto to get frontier-grade answers at a fraction of the per-task cost, while keeping the prompt cache warm across the whole session.
Does it hold up?
Too early to judge - all published scores are Unbiased's own preliminary runs and may change before final publication; no independent evaluations have appeared yet.
Built with Pareto 26.10 Preview
- Pareto 26.10 Preview (YFarmX analysis of pricing, release and benchmark claims)article · Details the Oct 1 release at $0.80/$3.20 per M tokens, explains the blend-instead-of-route mechanism, and flags that all scores are Unbiased's own preliminary runs.
- Unbiased Releases Pareto Multimodal AI Model - pricing and specs (TPS Report)article · Fact-checks the claims against confirmed data, noting the lack of independent benchmarks and production-ready telemetry for the new preview.
- Pareto 26.10 Preview (The Deploy Log - release ledger entry)article · Dated Oct 1 ledger entry tracking the OpenRouter listing ($0.80/$3.20, 1M context) as a verified AI deployment.
Learn more
First spotted on article: source.
More AI models
ElevenLabs 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 0.68GPT-Synopsys: OpenAI and Synopsys multi-year deal to build an AI model that operates EDA toolsOpenAI and Synopsys signed a multi-year agreement on Sep 30 to jointly develop GPT-Synopsys, a specialized model that…models · JEV 0.66
Get the week's best AI launches, plus 3 ideas worth building
One email every Saturday. Ranked by traction, not hype. Free.