DeepSeek · Efficient · Open weights
DeepSeek V4.1 Flash
DeepSeek's small, fast model on its new architecture, with native image understanding and a 1M-token window. It replaced V4 Flash on 10 September 2026; the old model names now route to it.
Model ID: deepseek-flash
What that means
A worked example
Per-token pricing is hard to feel. Take a moderately chatty production workload: 100 calls a day, roughly 10,000 tokens in and 2,000 out each time. On DeepSeek V4.1 Flash that comes to about:
$16 / month
Before caching, batching or volume discounts, all of which move this number a long way. Illustrative only.
Strong at
- -Very low cost
- -1M context
- -Open weights
Typical use
- -Bulk document processing
- -Cheap long-context tasks
- -Self-hosted deployments
Watch out for
- -Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC on weekdays
- -Consider data residency and governance before sending regulated data
Same tier
What else to look at
| Model | Developer | Context | In / 1M | Out / 1M |
|---|---|---|---|---|
| Claude Haiku 5.5 | Anthropic | 1M | $0.10 | $0.50 |
| Claude Haiku 4.5 | Anthropic | 200K | $1 | $5 |
| GPT-6 Luna | OpenAI | 1.05M | $0.10 | $0.50 |
| GPT-5.6 Luna | OpenAI | 1.05M | $0.20 | $1.20 |
| Gemini 3.5 Flash-Lite | 1.05M | $0.30 | $2.50 |
Verified 2026-10-11.