DeepSeek V4.1-Flash
Language Available Full comparison ↗ deepseek/deepseek-v4.1-flash · by DeepSeek
· mixture-of-experts
Pricing — 1 offering(s)
Input tokens
- $0.30 / 1M tokens (input) Current 2026-09-10 → present
Output tokens
- $1.20 / 1M tokens (output) Current 2026-09-10 → present
Off-peak input tokens
- $0.15 / 1M tokens (input) Current 2026-09-10 → present
Off-peak output tokens
- $0.60 / 1M tokens (output) Current 2026-09-10 → present
Cached input tokens (cache hit)
- $0.006 / 1M tokens (cached input) Current 2026-09-10 → present
Off-peak cached input tokens (cache hit)
- $0.003 / 1M tokens (cached input) Current 2026-09-10 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how DeepSeek V4.1-Flash fits into a cost-aware routing setup
See how →Capability profile
How DeepSeek V4.1-Flash rates across core capability dimensions, with the task-level evidence behind each rating.
1,000,000-token context, 384K maximum output (official pricing page).
Thinking and non-thinking modes. DeepSeek's change log reports GPQA Diamond 90.9 and HLE 36.8 (text-only subset); these are vendor-reported and not independently verified here.
Terminal-Bench 2.1 (89 real terminal tasks; Terminus 2 harness): completes 74.5% — fails roughly 1 task in 4 (high reasoning effort; run before the 2026-10-17 V4.1-Flash-1017 update). Terminal-Bench 4.0 (66 long-horizon multi-domain tasks; mini-swe-agent): 31.8% for the 1017 build the API now serves, up from 19.7% before the update. Independent (Vals AI). DeepSeek's change logs report Terminal-Bench 2.1 90.6 and DeepSWE v1.1 74.2 (2026-09-10) and Terminal-Bench 4.0 42.3 for 1017, vendor-reported on its own setup and above the independent figures.
Tool calls, the Responses API and the Anthropic-format API are all listed as supported on the official pricing page.
$0.30 / 1M cache-miss input and $1.20 / 1M output at peak, half that off-peak ($0.15 / $0.60), which covers roughly 80% of the week's hours. Cache hits are $0.006 peak / $0.003 off-peak. Official pricing page, checked 2026-09-28.
Ratings are estimated — limited independent data is available for this model.
Coding benchmarks
View the full coding benchmark leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| terminal-bench-2-1 | 74.5% 2026-09 | terminus-2 | 📊 leaderboard ↗ |
| terminal-bench-4-0 | 31.8% 2026-10 | mini-swe-agent | 📊 leaderboard ↗ |
Operator guidance
DeepSeek's current default model. Prefer it over V4-Pro unless you have your own evidence V4-Pro does better on your task: DeepSeek says V4.1-Flash surpassed V4-Pro on performance, cost and speed, but that's a vendor claim. Schedule batch-like work off-peak to pay half the headline rate.
Use cases
- Low-cost long-context and agentic work, especially when jobs can run off-peak (weekends, Chinese public holidays and outside 01:00-04:00 / 06:00-10:00 UTC on weekdays)
- Image-plus-text agent tasks on the DeepSeek API (V4-Pro has no vision support)
- Drop-in replacement for code still calling deepseek-v4-flash, which now routes here
Limitations
- All capability figures available so far are vendor-reported; no independent benchmark coverage checked yet, so ratings stay at unknown/moderate
- Architecture details (parameter counts, attention design) aren't published on the pages checked
- Peak-hour pricing is double the off-peak rate; costs depend on when requests run