Don't deploy big models locally, Mistral is launching a free API package for developers with direct domestic connectivity!

Do not access ONE-API and apply for multiple accounts as a staging service load! One is unethical and the other is blocked.

MistralAI launched the API free package has been a few days, has been hesitant to send out the reason is that Mistral has repeatedly fooled developers.... ...At the end of August it was originally said that it would not be available if you did not subscribe to the plan in September, but it turned out that most of the models were directly made available for free on September 17th.

Go directly to https://console.mistral.ai/ and go to the billing page to open it.

To join the free program you first have to be verified by phone, which I've done here. Using Google Virtual Number, domestic cell phone numbers are also supported for input, but I didn't try it.

After half a day of searching for any restrictions on the free package, I couldn't find any, and I finally came across a description of the limitations on the model types that can be used, as follows:

Mistral Two types of models are offered: free models and premium models.

It looks like free ones are available, request address format: https://api.mistral.ai/v1/chat/completions

Then just create the key and use it:

Note: Their api doesn't support the penalty parameter, you can get a worker to block it, otherwise it may be abnormal in many terminals.

Field worker scripts are not supported by mistral filtering (update cors cross-domain processing):

/**
* Welcome to Cloudflare Workers! This is your first worker.
*
* - Run "npm run dev" in your terminal to start a development server
* - Open a browser tab at http://localhost:8787/ to see your worker in action
* - Run "npm run deploy" to publish your worker
*
* Learn more at https://developers.cloudflare.com/workers/
*/


export default {
async fetch(request, env) {
// 处理预检请求
if (request.method === 'OPTIONS') {
return handleOptions(request);
}

const url = new URL(request.url);
url.host = 'api.mistral.ai';

let body;
try {
// 获取 body 数据并解析为 JSON
body = await request.json();
} catch (error) {
return new Response('Invalid JSON', { status: 400 });
}

// 定义支持字段列表
const allowedFields = ['model', 'messages', 'temperature', 'top_p', 'max_tokens', 'min_tokens', 'stream', 'stop', 'random_seed','response_format','tools','tool_choice','safe_prompt'];

// 过滤 body 中不在支持字段列表中的字段
const filteredBody = filterFields(body, allowedFields);

const newBody = JSON.stringify(filteredBody);
const modifiedRequest = new Request(url, {
method: request.method,
body: newBody,
headers: request.headers
});

return fetch(modifiedRequest);
}
}
// 过滤不支持参数
function filterFields(obj, allowedFields) {
let result = {};
for (const key in obj) {
if (allowedFields.includes(key)) {
result[key] = obj[key];
}
}
return result;
}
// 处理预检请求的函数
function handleOptions(request) {
const headers = {
'Access-Control-Allow-Origin': '*', // 允许所有来源
'Access-Control-Allow-Methods': 'GET, POST, OPTIONS', // 允许的方法
'Access-Control-Allow-Headers': 'Content-Type, Authorization', // 允许的头部
'Access-Control-Max-Age': '86400', // 预检请求的缓存时间（秒）
};

return new Response(null, {
status: 204,
headers: headers
});
}

The free code model requires a separate key to use

Although there are a lot of domestic price and low, good performance model (DeepSeek), and even a free big model API (Smart Spectrum), but the free should also be the best of the best.

Mistral has a specialized code model Codestral, the above content has been introduced, LMSYS ranking is quite high, you can access to Cursor try, use the feeling is okay.