Groq 404 model_not_found: mixtral-8x7b-32768 (shut down March 20, 2025, and its replacement was retired too)

Mixtral 8x7B was one of the first models on Groq and the id ended up in a great many tutorials from 2024. Groq emailed users on March 5, 2025 and removed it on March 20, 2025, with 15 days of notice. Eighteen months later the id is still in about 40,000 files on GitHub, and code that reaches it gets the same 404 model_not_found Groq returns for any unknown id. What makes this one awkward is the second step: the replacement Groq recommended, llama-3.3-70b-versatile, was itself shut down on August 16, 2026.

Last verified September 18, 2026 against Groq: deprecations · 3 min read

Provider	Groq (`api.groq.com/openai/v1`)
HTTP status	`404 Not Found`
Error code	`model_not_found`
Message	The model `mixtral-8x7b-32768` does not exist or you do not have access to it.
Announced / shut down	March 5, 2025 / March 20, 2025
Groq's replacements at the time	`mistral-saba-24b` or `llama-3.3-70b-versatile`
How common	About 40,000 files on GitHub still contain the id (code search, September 17, 2026)
Can you wait it out?	No

Short answer

At Groq: move to a current model; Groq's present recommendation for the 70B slot is openai/gpt-oss-120b or qwen/qwen3.6-27b.
If you already switched to llama-3.3-70b-versatile: that id works on our endpoint with a base URL change; see below.
Mixtral itself: the weights are public and some hosts still serve it, but none of the large OpenAI-compatible providers we track list it.

What the error looks like

HTTP 404 · Groq (api.groq.com/openai/v1)

{
  "error": {
    "message": "The model `mixtral-8x7b-32768` does not exist or you do not have access to it.",
    "type": "invalid_request_error",
    "code": "model_not_found"
  }
}

Groq returns this for retired ids and for ids that never existed. The deprecations page is the only place that tells the two apart.

The replacement chain

Step	Id	Status at Groq
What you pinned	`mixtral-8x7b-32768`	Shut down March 20, 2025 (announced March 5, 2025)
Groq's suggested replacement	`llama-3.3-70b-versatile`	Shut down August 16, 2026 (announced June 17, 2026)
Groq's other suggestion	`mistral-saba-24b`	Check Groq's model list; we do not track it
Groq's current suggestion for the 70B slot	`openai/gpt-oss-120b` or `qwen/qwen3.6-27b`	Active at Groq

Groq's notice periods have been short: 15 days for Mixtral, about two months for Llama 3.3 70B. If a model id is load-bearing for you, the deprecations page is worth a monthly look, or use our deprecation tracker, which lists Groq's entries with dates.

Fix 1: a current model at Groq

Change the id to one Groq serves today. Mixtral was an 8x7B mixture of experts with a 32K context; nothing on Groq is the same architecture, so re-run your evaluation prompts on the candidate. Groq's own suggestions for the slot are openai/gpt-oss-120b and qwen/qwen3.6-27b. Free-tier limits on Groq are covered on the 429 page.

Fix 2: keep llama-3.3-70b-versatile on a different base URL

Many projects made Groq's recommended change in 2025 and then broke again in August 2026. Llama 3.3 70B is an open-weight model; the id Groq used for it is accepted here as an alias for meta-llama/Llama-3.3-70B-Instruct-Turbo, so the migration is the base URL only:

Shell · the id Groq recommended in 2025 still works, on a different base URL

curl https://api.inferenceapis.com/v1/chat/completions \
  -H "Authorization: Bearer $INFERENCEAPIS_API_KEY" -H "Content-Type: application/json" \
  -d '{"model": "llama-3.3-70b-versatile", "messages": [{"role": "user", "content": "Say ok"}]}'

Same weights, different host: check the trust page for precision, region and retention, and the Llama 3.3 page for the details of that retirement.

Frequently asked questions

Can I get Mixtral 8x7B anywhere?

The weights are on Hugging Face and can be self-hosted. Among hosted OpenAI-compatible providers it has largely disappeared; we do not serve it and do not plan to.

Is mistral-saba-24b still on Groq?

We have not verified it. Groq's model list is the source of truth.

Where Inference APIs fits

We do not serve Mixtral 8x7B. We do serve the model Groq told Mixtral users to move to: llama-3.3-70b-versatile is an accepted id here and runs Llama 3.3 70B Instruct on GPUs in the United States, so a project that already made Groq's recommended change needs only a base URL change to keep working. Prices and measured speed are on the model page.

Create a free account API docs How a base-URL switch works

Something changed or wrong? Tell us and we will re-verify the entry.