// Halo
Use Mistral models pay-per-prompt — no subscription, no account
Europe's open-weight model family, served peer-to-peer. Fast, cheap, and priced per prompt — no subscription, no account.
Mistral is the European open-weight family, and its reputation is built on efficiency: a lot of quality per unit of cost and latency. That makes it a workhorse rather than a showpiece — the model you reach for when you have a lot of work to get through.
On Halo, independent operators serve Mistral alongside 140+ other models.
Priced for volume
You pay per prompt in USDC on Base, for the actual tokens used. No monthly plan, no minimum. For high-volume work, a cheap fast model on a pay-per-use price is a very different cost curve from a subscription — and you can route only the hard cases to something bigger.
No account, no API key
There’s no Mistral signup and no key to manage. Sign in with a social account, top up with card or crypto, and prompt.
Verifiable and private
Every result carries a statistical proof of execution, so a cheaper substitute can’t be passed off as the model you paid for. Prompts are encrypted to the operator, and confidential mode seals them inside a hardware enclave (TEE) the operator can’t read.
Mix and match
Pair it with a frontier model for the hard cases — see GPT-5 and Claude — or compare against the other open families, Llama and DeepSeek. If you’re building on this, tools and x402 covers the agent path.
Frequently asked questions
What is Mistral good for?
The line is known for a strong quality-to-cost ratio and low latency, which makes it a good default for high-volume or latency-sensitive work — classification, extraction, summarising, chat at scale — where a frontier model would be overkill.
Do I need a Mistral account?
No. Operators on Halo serve the models and you reach them through the network. There's no provider signup, no API key, and no subscription — you sign in with a social account and pay per prompt.
How is it priced?
Per prompt, in USDC on Base, for the tokens you actually use. Independent operators compete for each request, so prices tend to sit at the low end for a given model.
Can I mix it with other models?
Yes, and that's usually the smart pattern: a cheap fast model for the bulk of the work and a frontier model for the hard cases. Every model on the network sits behind the same sign-in, so switching is just picking a different one.