Mistral Large 4 vs Claude Fable 5.1: Ten Times the Output Price, One Public Licence

Claude Fable 5.1 costs $10.00 per million input tokens and $50.00 per million output, returns its first token in 3.70 seconds, and offers a 1M context with a 128K output ceiling. Mistral Large 4 lists at $1.36, $0.14 cached and $4.18 — twelve times cheaper on output — and is currently running at half those lines with no published end date. The cheapest version of this comparison is therefore 24× on output, and the most expensive version is 12×.

Neither multiple survives contact with the licences. Mistral's weights are public; Fable's are not, and that is a structural difference no discount closes.

The cards

• Mistral Large 4 — $1.36 in, $0.14 cached, $4.18 out; promoted at $0.68 / $0.07 / $2.09. 1.05 trillion parameters with 52 billion active, a 1.6-billion-parameter vision encoder, 1M context claimed and 524,288 measured. Weights public, 1,060.6 GB bf16, under the Mistral Ecosystem Open-Weight License.

• Claude Fable 5.1 — $10.00 in, $50.00 out; 1M context; 128K maximum output; 3.70 seconds to first token; no published weights.

Two columns are worth reading together. Mistral's output ceiling is four times Fable's, and its first token arrives slower — a trillion-parameter MoE is not built for latency. Fable's price buys responsiveness and a validated, widely measured model; Mistral's buys volume and an artefact you can run.

Per task, three ways

One million input tokens and 100,000 output tokens. Fable: $10.00 plus $5.00 — $15.00. Mistral at list: $1.36 plus $0.418 — $1.78. On promotion: 89 cents.

Fifteen times at list, seventeen times on promotion. On a larger generation — say one million output tokens — the gap widens to $50 against $4.18, and that is the workload where the difference stops being a line item and becomes a business model.

The licence is the asymmetry the price hides

Mistral's Ecosystem Open-Weight License permits commercial use, modification, distribution and further training, and reserves one business model: a model-as-a-service where third parties get meaningful control over inputs, parameters or training data needs a separate agreement. Embedded features, internal use and mere relaying are carved out.

Fable 5.1 has no such text because there is no file. That is not a criticism of Anthropic's product — it is a statement about what you are buying. A team that needs the checkpoint inside its own perimeter is not comparing prices, it is comparing availability, and only one side has an answer.

Who should pick which

If latency is the product, Fable's 3.70 seconds against a trillion-parameter model's unmeasured front-end is worth paying for, and no discount on a slower model changes the user's experience. If volume is the product — batch generation, document processing, long-form output — $4.18 against $50 is a decision that makes itself, provided you can wait.

And if you need ownership, the comparison ended two sections ago.

What to do with the promotion

Budget Mistral at $1.36 and $4.18. The promotional figures are real and currently active, but an end date nobody has published is the definition of a number you cannot plan on. The upside, if it persists, is a 24× output advantage rather than a 12× one — a pleasant surprise rather than a forecast.

Mistral Large 4 is not on OrcaRouter's catalogue, so this is not a route we can offer; it is callable through Mistral's own API and third-party hosts. Claude Fable 5.1 is on OrcaRouter at $10.00 and $50.00 with no markup, alongside 223 other models on one OpenAI-compatible endpoint — which makes one side of this comparison a routing rule and the other a separate integration.