Two models for generating answers and one for making decisions. Particle-1 focuses on coding efficiency, Mistral Large 4 combines reasoning with text and image understanding, and Joey-1 returns typed probabilities and scores.
Choose by the work you need done.
- Particle-1Post-trained on DeepSeek V4.1 Flash for software engineering. Particle reports a similar coding pass rate with fewer generated tokens in its own evaluation.particle/particle-1
- Mistral Large 4A sparse flagship preview for coding, document analysis and tools. Mistral documents native image understanding and a 1M context window.mistral/mistral-large-4
- Joey-1Baseten’s 2B decision model handles choices, yes/no questions and ordered scores without generating prose. It is not yet available on MiniRouter.baseten/joey-1
Check the route, then try a bounded task.
Use the model pages for current customer prices and setup. Particle-1 accepts text on MiniRouter; its publisher’s native image support is separate. Particle’s provider also applies peak-hour pricing, so compare total completed-task cost.
For coding, keep your tests and tools the same when comparing models. For document analysis, ask for citations to the material you supply and check them before acting.
Interested in classification or scoring? Read the decision API guide for currently supported models. Joey-1 uses a decision interface rather than Chat Completions.
Sources: Particle announcement, Mistral documentation and Vercel’s Joey-1 listing.