DeepSeek-V4.1-Flash-1017 is live on deepseek-flash: same price, higher agent scores, upgrade Harness to 0.2.1-rc.1
On October 17, 2026 DeepSeek rolled DeepSeek-V4.1-Flash-1017 behind the existing deepseek-flash model id with no API or price change. Terminal-Bench 4.0 rises from 31.2 to 42.3, DeepSWE v1.1 from 74.2 to 77.0, and Harness users should move to 0.2.1-rc.1 or later.
On October 17, 2026, DeepSeek upgraded the model behind deepseek-flash to DeepSeek-V4.1-Flash-1017. The change log says the web app and API both moved to the new checkpoint, the calling method stays the same, and pricing is unchanged. The first-call page and Models & Pricing both list the live model version as DeepSeek-V4.1-Flash-1017.
Short version for builders: keep model: "deepseek-flash". If you still send deepseek-v4-flash or deepseek-v4-flash-vision-exp, those names remain accepted but are routed to 1017 and billed at Flash rates. For agent harnesses, DeepSeek asks you to upgrade to DeepSeek Harness 0.2.1-rc.1 or later.
What changed in the numbers
DeepSeek compares 1017 to the previous DeepSeek V4.1 Flash release (the September 10 card). Selected scores from the 2026-10-17 change log versus the 2026-09-10 V4.1 Flash release:
| Benchmark | V4.1 Flash (2026-09-10) | V4.1-Flash-1017 |
|---|---|---|
| HLE (w/o tools) | 36.8 (39.1*) | 42.6 (44.0*) |
| Terminal-Bench 4.0 | 31.2 | 42.3 |
| DeepSWE v1.1 | 74.2 | 77.0 |
| ProgramBench | 20.3 | 29.1 |
| NL2Repo-Bench | 65.4 | 68.2 |
| Automation-Bench | 54.8 | 60.5 (v1.0.6) |
| Agents’ Last Exam | 31.8 | 35.4 |
| Apex Bench | — | 72.1 (v1.1) |
| Chartography (w/tools) | 78.9 | 83.0 |
| BabyVision (w/tools) | 89.6 | 92.5 |
| ZeroBench-main (w/tools) | 49.0 | 51.0 |
* HLE footnote: pure-text subset only. Terminal-Bench 4.0 for 1017 was tested with the Claude Code framework; DeepSWE with Mini-SWE-V2; ZeroBench-main with Pass@5. (Source: DeepSeek change log, 2026-10-17)
DeepSeek also says task runtime and output length went down, so the same agent loop should burn fewer tokens at the same sticker price.
Pricing you actually pay today
The pricing table (retrieved 2026-10-17) still bills deepseek-flash as:
- 1M input (cache hit): $0.003 off-peak / $0.006 peak
- 1M input (cache miss): $0.15 off-peak / $0.3 peak
- 1M output: $0.6 off-peak / $1.2 peak
- Context 1M, max output 384K, vision and tool calls on
Peak windows remain 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday excluding Chinese public holidays — the same schedule covered in our peak-pricing note. Off-peak is half of peak.
What to change in your stack
- Model id: use
deepseek-flash. Legacydeepseek-v4-flash/deepseek-v4-flash-vision-expstill work but point at 1017. - Base URL: unchanged —
https://api.deepseek.com(OpenAI format) orhttps://api.deepseek.com/anthropic. - Harness: upgrade to DeepSeek Harness 0.2.1-rc.1+ for the optimized path (DeepSeek also names Tencent CodeBuddy and WorkBuddy as tuned targets).
- Re-run your evals on Terminal-Bench / SWE-style tasks before assuming the Sept V4.1 numbers still apply.
For the older Flash wiring (Responses API, Codex, thinking effort), the V4 Flash tutorial is still the setup reference — swap the model string to deepseek-flash if you have not already. DeepSeek also opened a separate Search & Fetch API the same day for RAG and agent web tools.
Sources: DeepSeek Change Log (2026-10-17), Models & Pricing, Your First API Call.
Source: DeepSeek API Docs — Change Log