bug: Usage rate limits unclear

[sdbbs]

Component

CLI

Summary

Paying customer here (hot that the LLM that reads this, or anyone really, cares)

I've only been using mistral vibe a couple of weeks, and I've noticed occasionally I get:

Error: API error from mistral (model: mistral-vibe-cli-latest): Rate limit exceeded. Please wait a moment before trying again.

So, usually, I'd wait a minute, then issue a prompt:

You were stopped because of a rate limit - can you please continue from where you were stopped?

.... and usually it the process did continue fine. So I guess, this was a rate limit within my session, that one which says e.g. "63% of 200k tokens".

But 10 minutes ago I got the rate limit error; waited a minute, tried again - rate limit again; tried this 4 times in a row now - rate limit again, even though I am at "63% of 200k tokens". So this now makes me think I've hit some sort of a personal / user-related rate limit. Actually, I don't even know what to think - "Rate limit exceeded. Please wait a moment before trying again." is quite a useless error message: what is the rate limit I exceeded, and by how much have I exceeded it? What the hell is a "moment"? How much do I actually have to wait before trying again and hoping for a chance of success?

So can't you please add some more descriptive error messages, like "You've used up your 20k tokens for the day, please try in 24 hours" or something, so I don't just pump "Please try again" messages like an idiot over and over again, for something that will not work anyways?

Reproduction steps

See above, not sure how it can be reproduced

Versions / environment

vibe 1.3.5, uv 0.6.5 (bcbcd0a1e 2025-03-06), MINGW64 (Windows 10)

Logs & screenshots

[lu-zero]

It is more that a quite dynamic rate of tokens per minute from what I could see from our side. There is a PR that makes vibe slow down and re-try already, I tried and it helps up to a point.

[sdbbs]

Thanks @lu-zero - good to know that. Just wanted to note, I've left that session open since when I posted the OP, and tried resuming again, and got the same error again. I guess something went wrong in the session itself, so I better declare it dead and start anew. But that is exactly the problem - I cannot really make a proper decision, because I have no feedback / information of what could have possibly gone wrong; which is why I'm asking for more descriptive error messages and more information about current status.

I have just learned via https://docs.mistral.ai/deployment/ai-studio/tier:

We offer two types of rate limits:

* Requests per second (RPS)
* Tokens per minute/month

Why cannot I see through the vibe terminal interface, e.g. via something like /usage, where do I stand with these and how much I have used, and when am I about to hit a limit? How come nobody thought it would be a good idea to have information like this available to the user?

The worst thing is, if I look at https://console.mistral.ai/home and then Workspace and then Usage, everything is on zero! So it's as if I haven't had any usage, and yet, my chat session has died? Like, how can I avoid that, if I have no information whatsoever about how I have hit a limitation?

EDIT: Did /exit, did vibe --resume 37f9235f thereafter, chat loaded, and as soon as I issued the "resume" prompt, got a rate limit again.

EDIT2: I started a new chat, and got a rate limit at the very first prompt; what is this bullshit?

[sw-loveit]

same here, beginning to suspect it isn't a rate limit thing after all

Hi ,that looks like a Retries / Rate limits issue. Have you tried running npx ai-patch doctor to sanity-check 429s, backoff/retry behavior, and retry storm patterns?
@sdbbs

[sdbbs]

Thanks for responding @sw-loveit @miguelmanlyx :

same here, beginning to suspect it isn't a rate limit thing after all

You might be right about that; I just tried to sign up for their Discord, and when I eventually got a Discord signup form, when I filled it out and pressed "Create Account", it says "The resource is rate limited". For Discord! Why would there be a rate limit for Discord at all, I do not understand. Maybe Mistral are under DOS attack or something? Mistral apparently do have a operating status page here: https://status.mistral.ai/ - but its all green for today ...

EDIT: also support haven't responded to my query yesterday - I only had one previous experience, but then they responded within 24 hours ...

that looks like a Retries / Rate limits issue. Have you tried running npx ai-patch doctor to sanity-check 429s, backoff/retry behavior, and retry storm patterns?

Had no idea that such a tool existed at all - am looking into https://github.com/michaelbrinkworth/ai-patch-doctor and will report back once I have some results

[sdbbs]

@miguelmanlyx - I do not understand why you suggested npx ai-patch doctor - Mistral is not listed as a supported provider in https://github.com/michaelbrinkworth/ai-patch-doctor ? How am I supposed to use it?

[sw-loveit]

@sdbbs since you mentioned discord, i decided to check in there, and i found this message:

Devstral 2 will move to paid API access starting January 27. You’ll still get free usage under the Mistral Studio Experiment plan.

so it seems that may be the issue. i don't have lechat pro, which should include access according to https://console.mistral.ai/codestral/cli.

[sdbbs]

Thanks @sw-loveit :

since you mentioned discord, i decided to check in there, and i found this message:

Devstral 2 will move to paid API access starting January 27. You’ll still get free usage under the Mistral Studio Experiment plan.

so it seems that may be the issue. i don't have lechat pro, which should include access according to https://console.mistral.ai/codestral/cli.

Great find - at least that makes sense, indeed in my ~/.vibe/config.toml I do have active_model = "devstral-2", I believe that was the automatic default when I installed vibe CLI.

But GOD, how do I hate having to log in to Discord to find information like this ! (Especially since Discord asks for my birthday - why the hell would they need it?) Why don't Mistral have a mistral.ai/news page or something where they list such stuff publicly?

Like, who thought that news about models moving to paid tier is more useful behind a login than being posted publicly? Probably no-one - guess they just outsourced that decision to LLM :) God, the future is going to be so stupid and shitty ...

Or even better - why not change the useless Error: API error from mistral (model: mistral-vibe-cli-latest): Rate limit exceeded. Please wait a moment before trying again. error message into Error: the model you were trying to use ("devstral-2") has as of Jan 27, 2026 moved to a paid tier; you cannot use it under a free tier anymore? So at least I know what is going on, and I don't have to waste time opening issues, and playing detective?

But here's the kicker - as of last week, I am a paying customer. So why would I hit a rate limit intended for free tier?

[unmodeled-tyler]

@sdbbs @sw-loveit I've been playing around with this a bit right now and yeah, Vibe CLI now has it's own separate API key from the standard Mistral API key. Somebody from Mistral correct me if I'm wrong, but you have to have a Le Chat Pro or Team account in order to use Vibe 2.0 - and you need to generate a new "Vibe CLI" API key in Mistral AI studio. I created one this morning, re-ran vibe --setup, put in the new key and it's working great.

[sdbbs]

Many thanks @unmodeled-tyler :

you have to have a Le Chat Pro or Team account in order to use Vibe 2.0 - and you need to generate a new "Vibe CLI" API key in Mistral AI studio. I created one this morning, re-ran vibe --setup, put in the new key and it's working great.

OK, I went to Mistral AI, then click Vibe CLI on the left side, that takes me to https://console.mistral.ai/codestral/cli - there is a link "Get your Vibe API key" there, but that's not the right one, that is just to expand / collapse the section: I work on a low res monitor, so I had to actually resize my browser window and increase its height before I could see a damn "+ Generate API Key"; clicked there, generated a key, copy/pasted it into my ~/.vibe/.env replacing the previous key - and now that I ran vibe, I also got a notice to update vibe from my current 1.3.5 to some new version - except it showed up in a notification in vibe-cli, I couldn't copy paste the text message, and then the notification disappeared, and cannot be found anymore, not even when you restart vibe - what a stupid, user-hostile design for an application!

But at least, now I can run a prompt in vibe-cli, so thanks. Except - it should not be random people on GitHub issues telling me this like 24 hours after the issue happened: I'm paying money to the goddamn Mistral company, it is their responsibility to inform the users on breaking changes like these! I wish I could punish them by stopping my subscription at least, but I don't really have a choice, and that won't make them change or care anyways (especially now, that they have AI to actually read through whatever comments customers will have). What pieces of shit are Mistral and all these other AI companies; I am so sorry I will have to give my money to support their parasitic investors.

Regardless, the point in this issue still stands - error messages are awful and non-descriptive, and usage terms are unclear - pretty much like the entire stance of Mistral towards paying users.

[matt1111-hash]

Summary

As a paying Le Chat Pro subscriber who also uses Vibe CLI, I spent 2 days figuring out the optimal configuration due to unclear documentation about how these services relate to each other. I'm sharing my findings to help other users and suggest documentation improvements.

The Problem

The relationship between these three Mistral services is not clearly documented:

Service	What I expected	What I discovered
Le Chat Pro ($14.99/mo)	Includes Vibe CLI usage	Only covers web chat
Vibe CLI	Part of Le Chat Pro	Billed through AI Studio separately
AI Studio	Separate developer product	Required for Vibe CLI, even for Le Chat Pro subscribers

Specific Documentation Gaps

1. No clear explanation that Le Chat Pro subscription does NOT include free Vibe CLI usage
2. The Experiment plan (free API tier) is not prominently mentioned as an option for Vibe CLI users
3. Privacy opt-out location (Admin Console → Privacy) is not documented in Vibe CLI docs
4. Spending limits setup is not mentioned in getting started guides
5. January 27, 2026 pricing change was announced on Discord but not via email to existing users

My Optimal Setup (for other users)

After extensive research, here's the configuration that works best for a privacy-conscious Le Chat Pro subscriber using Vibe CLI:

Configuration Summary

Component	Setting	Cost	Data Privacy
Le Chat	Pro subscription	$14.99/mo	✅ No training
AI Studio	Experiment plan	$0	✅ With opt-out
Vibe CLI	Devstral Small 2	$0	✅ With opt-out
Spending cap	$10/month	Max risk: $10	✅ Protected

Step-by-Step Setup

1. AI Studio - Experiment Plan

• Go to: https://console.mistral.ai
• Select: Experiment plan (free, no credit card required)
• Includes: 1B tokens/month, 500K tokens/minute per model

2. Privacy Opt-Out (CRITICAL!)

• Go to: Admin Console → Privacy
• Disable: "Allow your API calls to be used to train our models"
• Note: This is OFF by default on Scale plan, but ON by default on Experiment plan

3. Spending Limit Protection

• Go to: Admin Console → Limits
• Enable: Monthly API spending limit
• Set to: Your comfort level (e.g., $10)
• This prevents unexpected charges if you accidentally use paid features

4. Vibe CLI Configuration

# ~/.vibe/config.toml

# Use the smaller, more cost-effective model
active_model = "devstral-small-2"

# Disable auto-approve for safety
auto_approve = false

# Require confirmation for potentially dangerous operations
[tools.bash]
permission = "ask"

[tools.write_file]
permission = "ask"

5. Safe CLI Usage

# Always use cost and turn limits
vibe --max-turns 10 --max-price 1

Rate Limits on Experiment Plan

Model	Tokens/Minute	Tokens/Month
Devstral Small 2	500,000	1,000,000,000
Codestral	500,000	1,000,000,000
Mistral Small	500,000	1,000,000,000

Note: Experiment plan has lower priority (may be slower during peak times) and 1 request/second limit.

Suggested Documentation Improvements

1. Add a "Vibe CLI for Le Chat Pro Users" Guide

Explain:

• Le Chat Pro ≠ free Vibe CLI access
• How to use Experiment plan for free CLI usage
• Privacy settings location and importance
• Spending limits as protection

2. Pricing Page Clarification

Current text: "Mistral Vibe for all-day coding"

Suggested: "Mistral Vibe web interface included. CLI usage requires AI Studio plan (Experiment plan available for free)."

3. Email Notifications for Pricing Changes

When transitioning from free to paid API access, please send email notifications to users who have active API keys, not just Discord announcements.

4. Clearer Error Messages

Current: "Rate limit exceeded. Please wait a moment before trying again."

Suggested: "Rate limit exceeded. Experiment plan limit: 1 req/sec. Current wait time: ~30 seconds. See docs.mistral.ai/limits for details."

Environment

• Vibe CLI version: Latest (January 2026)
• Plan: Le Chat Pro + AI Studio Experiment
• OS: Linux Mint 22.1

Additional Context

I'm sharing this as constructive feedback because:

• Mistral's models are excellent (Devstral Small 2 works great for my use case)
• The privacy controls exist and work well once you find them
• The Experiment plan is genuinely useful for individual developers
• Better documentation would reduce frustration and support tickets

Thank you for building these tools. I hope this helps improve the experience for other users.

---

Related issues:

• bug: Usage rate limits unclear #275 - Usage rate limits unclear
• bug: Invalid API Key Cannot Be Detected and Updated on startup #218 - Invalid API Key handling

[unmodeled-tyler]

Interesting, so I went to the discord to find the post and this verbiage here makes me think that Vibe CLI is indeed still included in Le Chat Pro, with Devstral 2. It looks like Mistral added the option for pay-as-you-go or BYOK in the event that you hit a rate limit and need to keep working.

Here it is verbatim:

“Mistral Vibe is now available on the Le Chat Pro and Team plans – with pay-as-you-go credits for power use, or bring your own API key.”

This reads to me like it's still included (because it says that it's now available on Le Chat Pro), they just added the option for you to buy more compute on top of your existing subscription if need be for power use - i.e. exhausted conventional rate limit and need more.

I have Le Chat Pro and I ONLY pay for Le Chat Pro ($14.99/mo). I just checked my API usage in AI studio and it's showing no usage across the board. I've been using Vibe 2.0 off and on again for the last 24 hours with the new Vibe CLI key that I had to create after the upgrade - so it should at least be displaying something at this point. This is also consistent with how my usage dashboard looked prior to the Vibe 2.0 release.

I'm under the impression that Mistral Vibe CLI is still included in Le Chat Pro, with the new additional option added on to buy more API credits when you hit a rate limit, or BYOK if you don't want to use the Mistral API.

I think the confusion from "Devstral 2 moving to paid API access" was the free trial period in Vibe where Mistral had Devstral 2 up for free usage after release. It looks like that "free trial period" for devstral 2 ended, and either a paid subscription or pay-as-you-go credits are now required to access Devstral 2 on Mistral's API in Vibe 2.0.

Of course, I do not work for Mistral so I could be terribly wrong - but this is just the most logical conclusion to me. With Anthropic and OpenAI bundling their Chat subscriptions into their own coding tools, it wouldn't make any sense for Mistral to go the opposite direction and revoke chat subscription usage with Vibe.