Inference APIs

404 model_not_found: llama-4-maverick and llama-4-scout (Groq)

Groq's two Llama 4 models were the only multimodal chat models on its free tier, which is why 404 The model `meta-llama/llama-4-maverick-17b-128e-instruct` does not exist or you do not have access to it tends to break screenshot, receipt and photo features rather than plain chat. Maverick was retired on 9 March 2026 and Scout on 17 July 2026. Groq's recommended replacement, openai/gpt-oss-120b, takes text only, so the fix depends on whether you were sending images.

Last verified November 18, 2026 against Groq deprecations, Groq model list · 6 min read

ProviderGroq
HTTP status404 Not Found
Error codemodel_not_found
MessageThe model `meta-llama/llama-4-maverick-17b-128e-instruct` does not exist or you do not have access to it.
Announced / shut downabout a month before each shutdown / 9 March 2026 (Maverick), 17 July 2026 (Scout)
When it happensSince the shutdown dates, on any self-serve request that names either model
Can you wait it out?No — the model was retired; the request fails until you change the model or the provider
Short answer
  • Fastest: change the model string to openai/gpt-oss-120b on Groq and raise max_tokens, because it reasons before answering. Different model, so re-check your prompts.
  • Least code change: keep meta-llama/llama-4-maverick-17b-128e-instruct and change only the base URL; Inference APIs accepts the id and answers with Qwen3-VL 235B.
  • So it does not happen again: read the model id from configuration and add a fallback for 404 and 429.

What happened

Groq's deprecations page lists two notices. meta-llama/llama-4-maverick-17b-128e-instruct was shut down for all self-serve tiers on 9 March 2026 with openai/gpt-oss-120b as the replacement. meta-llama/llama-4-scout-17b-16e-instruct, which the April 2025 wave had named as the replacement for the Llama 3.2 vision previews, followed on 17 July 2026 for free and developer tiers, again pointing at openai/gpt-oss-120b or qwen/qwen3.6-27b. Neither replacement accepts image input.

First, find where the id is pinned

The id is often in more places than the one that errored: a default in a settings file, an environment variable on the server, a second call for summarisation or titles, a framework default.

Find every copy of the id
grep -rn -E "meta-llama/llama-4-maverick-17b-128e-instruct" . --include='*.py' --include='*.ts' --include='*.js' --include='*.json' --include='*.yaml' --include='*.yml' --include='*.toml' --include='*.env*'
# and the places grep cannot see: environment variables on the server, a hosted secrets store, the provider dashboard's default model

Option 1 — switch to Groq's recommended replacement

If your requests were text only, change the string to openai/gpt-oss-120b and you stay on Groq's free tier. What is different:

  • It reasons before answering, so raise max_tokens or send reasoning_effort: "low"; a budget sized for Llama 4's direct answers can come back empty.
  • No images. A request with an image_url part is rejected. If you sent images, this option is not available and you need a host that still serves a vision model.
  • Context drops from 128K (Scout on Groq) to 131K, which is about the same, but Maverick users who relied on Groq's larger window should check their history trimming.
One string changes
resp = client.chat.completions.create(
    model="openai/gpt-oss-120b",   # was "meta-llama/llama-4-maverick-17b-128e-instruct"
    max_tokens=800,                 # reasoning counts against it
    messages=[{"role": "user", "content": "Hello"}],
)

Option 2 — keep the old id, change the base URL

Inference APIs accepts both Groq ids (meta-llama/llama-4-maverick-17b-128e-instruct, meta-llama/llama-4-scout-17b-16e-instruct, and the short forms without the meta-llama/ prefix) and serves them with Qwen3-VL 235B, an open vision-language model: images stay in the same image_url message parts, tool calling works, there is no thinking phase to budget for, and the window is 262,144 tokens, at $0.26 / $1.15 per million tokens.

Base URL and key change, model string stays
from openai import OpenAI

client = OpenAI(
    base_url="https://api.inferenceapis.com/v1",   # was https://api.groq.com/openai/v1
    api_key=os.environ["INFERENCE_API_KEY"],
)
resp = client.chat.completions.create(
    model="meta-llama/llama-4-maverick-17b-128e-instruct",   # the retired Groq id, accepted as an alias
    messages=[{"role": "user", "content": "Hello"}],
)

Qwen3-VL is a different model from Llama 4 and reads images differently; OCR and document-extraction prompts in particular deserve a re-run on your own samples. For text-only traffic, openai/gpt-oss-120b is served here as well at $0.20 / $0.80 and is the closer match to what Groq moved you to. Prices and capabilities: Qwen3-VL 235B here.

Option 3 — stop pinning one provider and one id

Put the model id in configuration and let a 404 or 429 fall through to a second provider. A retirement then degrades to a log line instead of an outage, and you choose when to migrate. LiteLLM's router does this in a few lines; the same shape exists in most frameworks.

LiteLLM proxy: Groq first, fall back on 404/429
# LiteLLM proxy config: Groq first, Inference APIs when Groq answers 404 or 429
model_list:
  - model_name: chat
    litellm_params:
      model: groq/openai/gpt-oss-120b
      api_key: os.environ/GROQ_API_KEY
  - model_name: chat-fallback
    litellm_params:
      model: inferenceapis/openai/gpt-oss-120b
      api_key: os.environ/INFERENCEAPIS_API_KEY

router_settings:
  fallbacks: [{ "chat": ["chat-fallback"] }]

The deprecations tracker lists every Groq retirement with its date and replacement, so you can see the next one coming.

Frequently asked questions

Does any provider still serve Llama 4 Maverick or Scout?

Several hosts still serve the open weights under Meta's ids. Groq's own ids are gone from its self-serve tiers. Inference APIs does not serve Llama 4; it accepts the ids and answers with Qwen3-VL 235B so that image features keep working.

Will my image requests fail on gpt-oss-120b?

Yes. GPT-OSS is text-only on every host, including Groq. Keep a vision-capable model for those requests and route text-only requests wherever you like.

Which Groq model takes images now?

At the time of writing Groq's production list has no multimodal chat model on the free tier; check the current list, because it changes every month or two. The deprecations tracker records each change with its date.

Where Inference APIs fits

Option 2 above keeps the old id working with a base-URL change: Inference APIs accepts meta-llama/llama-4-maverick-17b-128e-instruct and serves it with Qwen3-VL 235B at $0.26 / $1.15 per million tokens, billed per request with no daily cap. Groq's own replacement, openai/gpt-oss-120b, is served here too at $0.20 / $0.80.

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