← All updates

Mistral Ships Timestral Open Weights

Mistral's first time-series foundation models top GIFT-Eval, TIME and fev-bench, and run on a CPU

Hi!

DIT Model Radar tracks new AI models daily — real models or confirmed codenames only, no tools, no hype.

Mistral released Timestral, its first family of time-series foundation models, in four sizes from 5M to 209M parameters under a Modified MIT license. The 209M flagship ranks first among zero-shot models on GIFT-Eval, TIME and fev-bench, beating TimesFM-3 with fewer parameters, and needs no GPU at inference.

What changed: Timestral ships as four checkpoints (5M, 28M, 123M and 209M) that share one architecture and training recipe. Each reads up to 32,768 historical steps, forecasts up to 1,024 steps per pass across 21 quantiles, and takes multivariate targets plus past-only and known-future covariates. The 123M model comes within 0.2% of the 209M score at three quarters of its latency, and on an 8-thread server CPU a day-ahead forecast takes 10 ms on the 5M model and 140 ms on the 209M.

  • Open weights, Modified MIT license, free to self-host
  • Deploys on-prem, in a VPC or air-gapped; fine-tuning via Mistral Forge
  • No API pricing published

For forecasting teams, a zero-shot model that tops the public leaderboards and runs on a CPU makes the GPU budget optional.

Read more →


Top AI models. Lower costs.

Access GPT, Claude, Gemini, and more through one OpenAI-compatible API—and save up to 70% on AI API costs. Create your DIT account, get your API key, and start building.