404 model_not_found: moonshotai/kimi-k2-instruct (Groq)
Kimi K2 was the large agentic model on Groq's list, so 404 The model `moonshotai/kimi-k2-instruct` does not exist or you do not have access to it usually hits coding agents and tool-heavy pipelines. Groq retired the original id on 10 October 2025 in favour of moonshotai/kimi-k2-instruct-0905, then retired that one on 15 April 2026 with openai/gpt-oss-120b as the replacement. Apps that followed the first notice broke a second time.
Last verified November 18, 2026 against Groq deprecations, Groq model list · 5 min read
| Provider | Groq |
| HTTP status | 404 Not Found |
| Error code | model_not_found |
| Message | The model `moonshotai/kimi-k2-instruct` does not exist or you do not have access to it. |
| Announced / shut down | about a month before each shutdown / 10 October 2025 (kimi-k2-instruct), 15 April 2026 (kimi-k2-instruct-0905) |
| When it happens | Since the shutdown dates, on any self-serve request that names either id |
| Can you wait it out? | No — the model was retired; the request fails until you change the model or the provider |
- Fastest: change the model string to
openai/gpt-oss-120bon Groq and raisemax_tokens, because it reasons before answering. Different model, so re-check your prompts. - Least code change: keep
moonshotai/kimi-k2-instructand change only the base URL; Inference APIs accepts the id and answers with Kimi K2.7 Code. - 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 both notices: moonshotai/kimi-k2-instruct shut down 10 October 2025, replacement moonshotai/kimi-k2-instruct-0905; moonshotai/kimi-k2-instruct-0905 shut down 15 April 2026, replacement openai/gpt-oss-120b. Both applied to all self-serve tiers. Moonshot has since released Kimi K2.7, so the K2 line continues, just not on Groq.
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.
grep -rn -E "moonshotai/kimi-k2-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 modelOption 1 — switch to Groq's recommended replacement
One string changes and you stay on Groq. What is different about openai/gpt-oss-120b:
- It reasons before answering; Kimi K2 Instruct did not. Raise
max_tokensor sendreasoning_effort: "low". - It is a smaller model (117B total, 5B active, against K2's 1T total, 32B active). Long agent runs and large tool schemas that K2 handled may need more retries.
- Context drops to 131K from K2's 262K on Groq.
resp = client.chat.completions.create(
model="openai/gpt-oss-120b", # was "moonshotai/kimi-k2-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 moonshotai/kimi-k2-instruct, moonshotai/kimi-k2-instruct-0905 and the short forms as aliases and serves them with Kimi K2.7 Code, Moonshot's current K2 release: the same family and tool-calling habits, a 262,144-token window, at $0.89 / $4.42 per million tokens with cached input at $0.18.
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="moonshotai/kimi-k2-instruct", # the retired Groq id, accepted as an alias
messages=[{"role": "user", "content": "Hello"}],
)K2.7 Code thinks before answering by default and returns a reasoning_content field, which K2 Instruct never did. Send reasoning_effort: "none" to get the old direct-answer behaviour, and expect it to be slower per token than Groq's hardware was. For a cheaper text-only agent model, openai/gpt-oss-120b is served here at $0.20 / $0.80. Prices and capabilities: Kimi K2.7 Code 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 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
Is Kimi K2.7 Code the same model as kimi-k2-instruct?
No. It is the newest release of the same family, tuned for coding and long agent runs. Prompts written for K2 Instruct work, but outputs differ and thinking is on unless you switch it off.
Why is Groq's replacement a different family?
Groq's list favours models that run well on its hardware. GPT-OSS 120B is small, fast and open, which fits; K2's trillion parameters do not.
Where else is Kimi K2 served?
Moonshot's own platform serves the K2 line, and several open-model hosts serve the weights. The Groq ids are only accepted here, where they alias to K2.7 Code.
Where Inference APIs fits
Option 2 above keeps the old id working with a base-URL change: Inference APIs accepts moonshotai/kimi-k2-instruct and serves it with Kimi K2.7 Code at $0.89 / $4.42 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.
