Industry News
NVIDIA NIM Removes DeepSeek V4 and Mistral Medium Models from Free API
NVIDIA removed DeepSeek V4 Flash, DeepSeek V4 Pro and Mistral Medium 3.5 128B from the NIM serverless API on 7 August 2026. We confirmed the removals against the live endpoint and tracked where these models still run.
Key takeaways
- check_circleNVIDIA NIM removed deepseek-ai/deepseek-v4-flash, deepseek-ai/deepseek-v4-pro and mistralai/mistral-medium-3.5-128b on 7 August 2026.
- check_circleThe removals were confirmed against the live /v1/models endpoint, not picked up from rumour or secondary sources.
- check_circleDeepSeek V4 Flash and V4 Pro still run on several providers. Mistral Medium 3.5 128B currently has no other verified route in our index.
What happened
On 7 August 2026 at 09:08 UTC, the AZ Labs model index monitor checked the NVIDIA NIM serverless API endpoint and found three models missing from the listing. The check returned HTTP 200, which means the endpoint was healthy and the models really are gone from the catalogue.
The removed models were deepseek-ai/deepseek-v4-flash, deepseek-ai/deepseek-v4-pro and mistralai/mistral-medium-3.5-128b. All three had been listed since 21-22 July 2026 as part of NVIDIA's free developer tier for prototyping.
What these models were
DeepSeek V4 Flash and V4 Pro are DeepSeek's open coding models, served by NVIDIA NIM alongside models from other vendors. Mistral Medium 3.5 128B is Mistral's mid-size 128-billion-parameter model. On NIM all three ran under the free developer/evaluation tier, which NVIDIA describes as free access for unlimited prototyping and free serverless APIs for development.
Our health probes showed V4 Pro and Mistral Medium returning OK as recently as 1 August. V4 Flash's last confirmed OK was 21 July, and its final probe on 1 August failed with HTTP 529, which suggests the origin was already struggling before the listing disappeared.
Where these models still run
A removal from NVIDIA NIM does not remove the models from the market. DeepSeek V4 Flash and V4 Pro remain available through the DeepSeek official API, OpenRouter, OpenCode Go, Cline and Ollama Cloud. V4 Flash also still has a free window on OpenCode Zen and a hosted copy on Hugging Face.
Mistral Medium 3.5 128B is the one to watch. Our index has no other verified route for it, so anyone who relied on the NIM route should check Mistral's own API or an alternative hosting option before building on it.
What this means for builders
Free model tiers change without announcement. NVIDIA has added and removed models repeatedly this year. On 5 August alone, nvidia/nemotron3-ultra-genrm appeared in the catalogue and was gone by the afternoon. Treat any free serverless route as a test surface, not a production dependency.
If you call integrate.api.nvidia.com directly, pin a fallback route and re-test before you need it. Our free-model health checks now run weekly against every verified free route, so availability changes get caught quickly.
Frequently asked questions
Is the removal permanent?
Unknown. NVIDIA changes the catalogue without announcements. The monitor keeps checking, and a model that returns will be recorded as added.
Were these paid models?
No. All three were free developer/evaluation tier on NVIDIA NIM, which NVIDIA describes as free access for unlimited prototyping.
Does this affect apps using DeepSeek or Mistral elsewhere?
Only if they pointed at the NVIDIA NIM routes. The models themselves still run on other providers.