Inference APIs

404 model_not_found: deepseek-r1-distill-llama-70b (Groq)

DeepSeek R1 Distill Llama 70B was the first reasoning model on Groq's free tier and the one most "thinking" demos from early 2025 were built on, which is why 404 The model `deepseek-r1-distill-llama-70b` does not exist or you do not have access to it tends to break code that parses <think> blocks. Groq retired it on 2 October 2025 for all self-serve tiers. The -specdec variant and the Qwen 32B distill had already gone in the April 2025 wave.

Last verified November 18, 2026 against Groq deprecations, Groq model list · 5 min read

ProviderGroq
HTTP status404 Not Found
Error codemodel_not_found
MessageThe model `deepseek-r1-distill-llama-70b` does not exist or you do not have access to it.
Announced / shut downabout a month earlier, by email and on the deprecations page / 2 October 2025
When it happensSince 2025-10-02, on any self-serve request that names this 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 deepseek-r1-distill-llama-70b and change only the base URL; Inference APIs accepts the id and answers with DeepSeek V4 Flash.
  • 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 the notice: shut down 2 October 2025, all self-serve tiers, replacements llama-3.3-70b-versatile or openai/gpt-oss-120b. The first of those was itself retired on 16 August 2026, leaving openai/gpt-oss-120b as the standing recommendation. DeepSeek's own line has moved on to V4, which reasons natively rather than through a distilled Llama.

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 "deepseek-r1-distill-llama-70b" . --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

One string changes and you stay on Groq's free tier. What is different about openai/gpt-oss-120b:

  • Reasoning arrives in a separate field (reasoning on Groq) instead of inline <think> tags in content. Code that stripped the tags can be deleted; code that depended on them needs the field instead.
  • Effort is adjustable with reasoning_effort (low, medium, high); the distill had one setting.
  • It is a different base model; the distill's Llama-flavoured answers and GPT-OSS's differ in tone, so re-run your test set.
One string changes
resp = client.chat.completions.create(
    model="openai/gpt-oss-120b",   # was "deepseek-r1-distill-llama-70b"
    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 deepseek-r1-distill-llama-70b (and the -specdec id) as an alias and serves it with DeepSeek V4 Flash, DeepSeek's current reasoning model: thinking on by default, returned in a reasoning_content field, a 1,048,576-token window, tool calling and JSON output, at $0.19 / $0.38 per million tokens with cached input at $0.04. It is cheaper than the distill was on any paid host.

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="deepseek-r1-distill-llama-70b",   # the retired Groq id, accepted as an alias
    messages=[{"role": "user", "content": "Hello"}],
)

V4 Flash puts its reasoning in reasoning_content, never in <think> tags inside content, so tag-stripping code becomes a no-op and tag-parsing code must read the field. reasoning_effort: "none" switches thinking off when you only want the answer. Groq's replacement openai/gpt-oss-120b is served here too at $0.20 / $0.80. Prices and capabilities: DeepSeek V4 Flash 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

My code parsed <think> tags. What do I change?

Read usage.completion_tokens_details.reasoning_tokens for the count and the message's reasoning_content field for the text. Both DeepSeek V4 Flash here and GPT-OSS on Groq keep reasoning out of content.

Is DeepSeek V4 Flash the same as R1?

No. R1 Distill Llama 70B was a Llama 3.3 fine-tuned on R1's reasoning traces. V4 Flash is a native DeepSeek model from 2026, stronger on the same benchmarks, with a window eight times larger. Prompts carry over; expect different wording.

Does the distill still exist anywhere?

The open weights (deepseek-ai/DeepSeek-R1-Distill-Llama-70B) are still hosted by some providers under their own ids. Groq's id is accepted only here, where it aliases to V4 Flash.

Where Inference APIs fits

Option 2 above keeps the old id working with a base-URL change: Inference APIs accepts deepseek-r1-distill-llama-70b and serves it with DeepSeek V4 Flash at $0.19 / $0.38 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.