Inference APIs
Reference/Errors/OpenRouter

404 "No endpoints found" (OpenRouter): for this model, matching your data policy, or that support tool use

Three different problems share the 404 No endpoints found wording on OpenRouter, and the fix differs for each. OpenRouter's own error reference documents the envelope ({"error": {"code": 404, "message": "..."}}) and the general not_found case; the three messages below are what users report in opencode, aider, LibreChat and the OpenRouter community, so treat the exact wording as current-as-reported rather than guaranteed. Together they are the most common reason an OpenRouter call to a free or niche open model fails outright.

Last verified November 20, 2026 against OpenRouter error reference, OpenRouter provider routing, opencode issue #1002 (tool use variant), aider issue #2444 (retired slug variant) · 5 min read

ProviderOpenRouter
HTTP status404 Not Found
Messages (as reported by users)No endpoints found for <model>. · No endpoints found matching your data policy. · No endpoints found that support tool use. To learn more about provider routing, visit: https://openrouter.ai/docs/provider-routing
When it happensThe routing step finds no provider endpoint for the model after applying your account's privacy settings, provider preferences and the request's required features
Can you wait it out?Sometimes: a free endpoint that is overloaded or removed can come back. A retired slug or a data-policy mismatch does not
Short answer
  • "for <model>": the slug has no live endpoint, usually a retired or renamed model (deepseek/deepseek-coder, deepseek-chat-v3:free). Check the slug on openrouter.ai/models.
  • "matching your data policy": your privacy settings exclude every provider that serves it. Adjust at openrouter.ai/settings/privacy or pick another model.
  • "that support tool use": none of the model's providers accept tools; common on :free DeepSeek and Kimi endpoints. Drop tools, pick a paid endpoint, or run the model elsewhere.

"No endpoints found for <model>": the slug is dead

OpenRouter removes a model when its last provider drops it, and renames slugs when a new version lands. A pinned deepseek/deepseek-coder (aider, 2024) or deepseek/deepseek-chat-v3:free (2025) then returns this 404 while the request itself is fine. The :free suffix is the usual culprit: free endpoints appear and disappear with provider capacity, and the message reads the same whether the slug never existed or stopped an hour ago.

Fix: look the model up on openrouter.ai/models and copy the current slug, or call GET https://openrouter.ai/api/v1/models and search the list. In LiteLLM, aider and opencode the slug is prefixed openrouter/; the part after the prefix must match exactly.

Check the slug
curl -s https://openrouter.ai/api/v1/models | python3 -c "import json,sys; ids=[m['id'] for m in json.load(sys.stdin)['data']]; print([i for i in ids if 'deepseek' in i])"

"No endpoints found matching your data policy": a setting, not the model

OpenRouter lets you forbid providers that train on prompts or that are outside a region, and lets you pin an allowed-provider list. If every provider that serves the model is excluded by those settings, the request has nowhere to go and this 404 is the answer. It commonly appears on new models that only one provider serves, and on accounts where an allowed-providers list was set months ago and forgotten.

Fix: open Settings → Privacy and either relax the training/data policy or clear the ignored and allowed provider lists; or choose a model served by a provider your policy accepts. Per request, the provider object can also override preferences (allow_fallbacks, order, data_collection); see the provider routing docs.

"No endpoints found that support tool use": the model is served, tools are not

When the request carries tools, OpenRouter only routes to endpoints that advertise tool support. Free DeepSeek R1/Chimera and Kimi K2 endpoints often do not, so an agent framework that always sends a tool list (opencode, Cline, LangChain agents) fails on exactly the models hobbyists pick. The opencode issue tracking it was closed as not fixable on the client side; the workarounds were to mark the model as having no tool calling or to use a paid endpoint.

Fix: send the request without tools if the task allows; set provider: {"require_parameters": true} and accept the paid endpoint it picks; or choose a model whose endpoints support tools, which the model page on OpenRouter lists per provider.

Force an endpoint that honours the request
resp = client.chat.completions.create(
    model="moonshotai/kimi-k2.6",          # not the :free variant
    messages=[{"role": "user", "content": "What is the weather in Paris?"}],
    tools=TOOLS,
    extra_body={"provider": {"require_parameters": True, "allow_fallbacks": True}},
)

When the open model is served nowhere on OpenRouter

All three variants come from the same design: OpenRouter is a router over other hosts, so a model is available only when some host serves it with the features and policy you need. For the widely used open models you can skip the router and call a host that serves them directly, with tool calling on and one published price. Inference APIs serves DeepSeek V4 Pro and Flash, Kimi K2.6 and K3.1 Pro, GPT-OSS 120B and 20B, Qwen3.8 Flash, GLM 5.3 and Llama 3.3 70B with the same OpenAI chat request; there is no free tier, so there are also no free endpoints to disappear.

Python
from openai import OpenAI
import os

client = OpenAI(base_url="https://api.inferenceapis.com/v1", api_key=os.environ["INFERENCEAPIS_API_KEY"])
resp = client.chat.completions.create(
    model="moonshotai/Kimi-K2.6",   # $0.98 / $4.55 per 1M tokens, tools supported
    messages=[{"role": "user", "content": "What is the weather in Paris?"}],
    tools=TOOLS,
)

Frequently asked questions

Is 404 No endpoints found the same as a plain 404 Not Found?

No. A bare {"message": "Not Found", "code": 404} means the slug never existed (usually a typo); "No endpoints found" means the slug resolves but routing found no endpoint under your settings and request.

Why does it work in the OpenRouter chat page but not in my app?

The chat page sends no tools and uses the account's default settings; your app may send a tool list, a provider preference or a required parameter that excludes every endpoint.

Do free endpoints come back?

Often, but on no schedule. If the app matters, do not pin a :free slug without a paid fallback.

Where Inference APIs fits

Option 3 below is for the open models: when OpenRouter has no endpoint for the model, or none that supports tools, the same OpenAI-style request runs against Inference APIs, which serves DeepSeek V4, Kimi K2.6 and K3.1 Pro, GPT-OSS, Qwen3, GLM and Llama 3.3 70B with tool calling on, at one published price each.

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