DeepSeek V4.1 Flash is now available on PrimaLabs serverless. It is one of the most requested models from developers building on our platform, and you can call it today with the same API key and endpoint you already use.
Why this model matters
DeepSeek built V4.1 Flash to make input-heavy agentic workloads cheaper to run: agents and pipelines that read far more than they write.
- Mixture of experts: 552B total parameters, 8B active per token during prefill and 16B during decode
- One-million-token context window, with a global KV cache of 890 bytes per token, which DeepSeek puts at about a quarter of V4 Flash’s
- Text input, text output
- Adjustable reasoning effort, 1 to 100, to trade depth for latency per request
- MIT license
What it’s good at
DeepSeek’s results at maximum reasoning effort (self-reported, not independently verified):
| Benchmark | Score | Domain |
|---|---|---|
| DeepSWE v1.1 | 74.2 resolved | Software engineering |
| Codeforces | 3471 rating | Competitive programming |
| CyberGym | 88.1 | Security |
| GPQA Diamond | 90.9 | Science reasoning |
| HLE with tools | 63.9 | Tool use |
On Artificial Analysis, V4.1 Flash (Max, reasoning) scores 39 on the Intelligence Index, #9 of 120 models.
Get started
Model ID: primalabs-ai/DeepSeek-V4.1-Flash
curl -sS https://api.primalabs.ai/v1/chat/completions \
-H "Authorization: Bearer <PRIMALABS_API_KEY>" \
-H "Content-Type: application/json" \
-d '{"model":"primalabs-ai/DeepSeek-V4.1-Flash",
"messages":[{"role":"user","content":"Hello"}]}'Pricing
Introductory pricing: 20% off list for a limited time. USD per 1M tokens.
| Token type | List | Introductory |
|---|---|---|
| Input | $0.24 | |
| Cached input | $0.0048 | |
| Output | $0.96 |
Cached input is metered and billed separately from fresh input, so an agent loop that resends the same system prompt pays the cached rate on most of its turns.
Moving from DeepSeek V4 Flash
We are retiring DeepSeek V4 Flash at 11:59 PM PT on Wednesday, October 21, 2026. To switch, change the model ID to primalabs-ai/DeepSeek-V4.1-Flash. If you need V4 Flash after that date, email [email protected].
Try it on your workload
Benchmarks tell you where to start; your traffic tells you the answer. Get an API key and run V4.1 Flash on your own prompts, or talk to an engineer about dedicated capacity and rate limits for production. Join us on Discord.
Notes: Specs and benchmark scores from DeepSeek’s model card (huggingface.co/deepseek-ai/DeepSeek-V4.1-Flash); DeepSeek’s scores are self-reported at maximum reasoning effort. Intelligence Index from artificialanalysis.ai. Results vary by workload.