Primary use stated or inferred cautiously from official documentation.
M
MISTRAL AI / mistralai/Timestral-209M
Timestral 209M
Mistral numerical model family released on October 15, 2026 in four sizes: 5M, 28M, 123M and 209M. It performs zero-shot univariate and multivariate probabilistic forecasting. It accepts history, past covariates and known future covariates; it neither chats nor generates text.
01 / SUMMARY
The essential
Maximum input capacity when published by the source.
Documented maximum generation limit.
Verified availability channels.
Primary functional family and verified specialties.
Declared terms for API access, model weights, or self-hosting.
02 / TECHNICAL SHEET
Limits and integration
| API ID | mistralai/Timestral-209M |
|---|---|
| Model type | Time series |
| Access model | Restricted weights |
| License | Modified MIT |
| Deployment | CPU / GPU |
| Release | 15 OCT 2026 |
| Knowledge cutoff | Not published |
| Entrance | Numerical data |
| Exit | Numerical data |
| Context window | 32,768 time steps |
| maximum output | 1,024 steps per pass · 21 quantiles |
| Reasoning | Not published |
| Published tools | |
| Structured Outings | Not published |
| Batch processing | Not published |
| Prompt cache | Not published |
| Fine-tuning | Not published |
| Verified platforms | PyTorch · Hugging Face |
Forecasting sensors, demand and capacity. It processes up to 32,768 historical time steps and predicts 1,024 steps per pass, with 21 quantiles. Official code supports CPU or GPU inference and evaluation on your own data.
Limits are time steps, not tokens. Predictions do not guarantee future outcomes. Avoid future-data leakage during validation and check distribution shifts. The Modified MIT license restricts rights when consolidated monthly revenue exceeds USD 20 million.
03 / CAPABILITIES
What it can do
Orientation
Forecasting sensors, demand and capacity. It processes up to 32,768 historical time steps and predicts 1,024 steps per pass, with 21 quantiles. Official code supports CPU or GPU inference and evaluation on your own data.
Context
32,768 time steps · 1,024 steps per pass · 21 quantiles
Tools and integration
Not published
Access
Hugging Face
Input modalities
Numerical data
04 / PRICES
Documented cost
| Concept | Worth | Unit/condition |
|---|---|---|
| Standard input | No published official API rate | Standard API rate |
| Cached input | Not published | Reading reused prefixes |
| Cache write or storage | Not published | The condition varies by provider |
| Standard output | Self-hosted infrastructure cost | May include reasoning tokens |
| Batch input | Not published | Asynchronous processing |
| Batch output | Not published | Asynchronous processing |
No public API tariff has been verified for these checkpoints. Downloads do not cover infrastructure or maintenance. License restrictions also apply to derivatives: review the official text before commercial deployment.
i Prices change and may depend on level, region or context length. Check the source before making a decision.
05 / EVALUATIONS
How to read the results
“
Metrics reported by Mistral for each checkpoint, not reproduced by Inferama. Lower MASE and CRPS are better; higher fev-bench skill scores are better. These are not accuracy percentages or conversational tests. The technical report is still pending.
| Benchmark | Result | Metric | Source |
|---|---|---|---|
| GIFT-Eval MASE | 0,658 | Lower is better | View source ↗ |
| GIFT-Eval CRPS | 0,446 | Lower is better | View source ↗ |
| TIME MASE | 0,626 | Lower is better | View source ↗ |
| TIME CRPS | 0,522 | Lower is better | View source ↗ |
| fev-bench SQL skill score | 50,39 | Higher is better | View source ↗ |
| fev-bench MASE skill score | 39,49 | Higher is better | View source ↗ |
Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.
06 / VERSIONS
Chronology
Timestral 209M
Version added to Inferama's verified catalog.
07 / SOURCES