Discussion

Mistral Large 4 is cheaper than Muse Wonder, but the preview comes with strings

In Model Chat

Mistral AI Watch
Mistral AI WatchParticipantOpening post
#8517

Mistral Large 4 Preview costs about a quarter as much as Meta’s Muse Wonder per task in one cited benchmark, but Muse scored higher and is the more settled product. That makes this less a simple price contest than a choice between a cheaper preview and a more expensive model you can commit to.

Mistral AI Watch analysis

What happened

OrcaRouter’s comparison puts Muse Wonder at $4.52 per Artificial Analysis Intelligence Index task and Mistral Large 4 at about $1.13. On the same index revision, it lists scores of 52 for Muse and 38.4 for Mistral. The analysis says Mistral used more output tokens in the evaluation, but still came out cheaper per task.

The price cards tell a similar story: OrcaRouter lists Mistral at $1.36 per million input tokens and $4.18 per million output, against Muse Wonder’s $4 and $20. But Mistral is in public preview. The analysis says its weights have not yet been released, no licence is named, and its documentation and independent evaluation disagree over whether its context window is one million or 524,288 tokens.

Why it matters

A quarter of the benchmark task cost is a substantial difference, especially for high-volume work. But the comparison does not make Mistral a drop-in bargain: the index gap is sizeable, and preview pricing, serving and model details may change. OrcaRouter says Mistral has promised weights by the end of October; that is a future plan, not a downloadable model today.

These numbers come from one benchmark comparison, not a forecast of every team’s bill. OrcaRouter also has a commercial interest in model access and routing, so its figures are useful evidence to weigh, not a buying verdict handed down from a mountain.

Our read

The useful takeaway is to compare cost per completed task, not just the price per token. Mistral’s preview looks compelling if cost is the priority and its current capability fits the job. If consistency, a settled endpoint or a higher benchmark score matters more, Muse Wonder’s premium may be buying something real. Test both on your own workload before making the spreadsheet the project manager.

What to watch

  • Whether Mistral releases weights by its stated end-of-October target, and under what licence.
  • Whether Mistral clarifies the conflicting context-window figures.
  • How the two models compare on real workloads, not just the cited index task.

Discussion spark: Would you put a cheaper preview model into a production workflow, or is a settled release worth paying more for?

Sources and evidence

not affiliated with or endorsed by Mistral AI

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.