DeepSeek V4.1 Flash Guide
On 2026-09-10 DeepSeek officially released DeepSeek-V4.1-Flash, the smallest model in its new architecture family. Prices dropped sharply, and DeepSeek states that V4.1 Flash now outperforms V4 Pro across performance, cost, speed and total time. This page covers pricing, specs and benchmarks verified only against DeepSeek's official pricing page and change log, plus setup in VS Code GitHub Copilot Chat and exactly what happens to V4 Pro from 2026-09-14. See also the pricing comparison, the token calculator and the release timeline.
The short version
- V4.1 Flash is the current workhorse. The model ID you pass in API requests is
deepseek-flash(labels and options inside VS Code may differ as the extension is updated). 1M context, 384K max output, native image input. - Prices fell sharply versus V4 Pro. Input (cache miss) $0.15 off-peak / $0.30 peak; output $0.60 off-peak / $1.20 peak, in USD per 1M tokens.
- From 13:00 JST (04:00 UTC) on 2026-09-14, requests to
deepseek-v4-proare routed to V4.1 Flash and billed at the V4.1 Flash price. DeepSeek has not announced a V4.1 Pro release date.
Official pricing (as of 2026-09-11)
USD per 1M tokens. Off-peak rates are half the peak rates.
| Item | Off-peak | Peak |
|---|---|---|
| Input (cache hit) | $0.003 | $0.006 |
| Input (cache miss) | $0.15 | $0.30 |
| Output | $0.60 | $1.20 |
📘 cache hit vs cache miss: if the beginning of your prompt is still in DeepSeek's cache it counts as a cache hit (input $0.003 per 1M tokens — 1/50 of cache miss), otherwise it counts as a cache miss ($0.15). Sending the same long system prompt in the same order makes hits more likely.
Unit price vs V4 Pro (reference)
| Item | V4 Pro (through 2026-09-13) | V4.1 Flash | Change |
|---|---|---|---|
| Input cache miss (off-peak) | $0.66 | $0.15 | ~77% lower |
| Input cache hit (off-peak) | $0.022 | $0.003 | ~86% lower |
| Output (off-peak) | $1.98 | $0.60 | ~70% lower |
⏰ Peak hours: 01:00-04:00 and 06:00-10:00 UTC, Monday-Friday only (JST 10:00-13:00 / 15:00-19:00; Beijing 09:00-12:00 / 14:00-18:00). Weekends are off-peak all day. The peak/off-peak scheme took effect 2026-08-16 16:00 UTC. Source: api-docs.deepseek.com (retrieved 2026-09-11).
Estimate your own cost with the token calculator, which lists both off-peak and peak rows for V4.1 Flash. For cross-provider comparison see pricing (DeepSeek).
What happens to V4 Pro (retirement schedule)
Only facts stated in DeepSeek's official documentation.
⚠️ Behaviour from 13:00 JST on 2026-09-14
- DeepSeek states that V4.1 Flash "has comprehensively surpassed V4 Pro in performance, cost, speed, and total time, so we plan to retire V4 Pro in an orderly manner."
- From 12:00 Beijing time on 2026-09-14 (= 04:00 UTC = 13:00 JST), and until V4.1 Pro is released, all requests to
deepseek-v4-proare routed to V4.1 Flash and billed at the V4.1 Flash price. - In other words, your keys and code keep working — but the model actually answering and the rate you pay become V4.1 Flash.
- No V4.1 Pro release date has been announced. The docs only say "until V4.1 Pro is released in the future".
- Because the served model changes, re-check responses in production once after the cutover.
Legacy model names
| Model name | Status | Actual behaviour |
|---|---|---|
deepseek-flash | Current | DeepSeek-V4.1-Flash. Use this for new code |
deepseek-v4-pro | Retiring | Routed to V4.1 Flash and billed at the Flash price from 13:00 JST on 2026-09-14 |
deepseek-v4-flash | Retired | Accepted temporarily for compatibility; served by V4.1 Flash at the Flash price |
deepseek-v4-flash-vision-exp | Retired | Same as above (the old vision experimental model folded into V4.1 Flash) |
deepseek-chat / deepseek-reasoner | Discontinued | Legacy aliases ended on 2026-07-24 |
Source: Models & Pricing notes (1) and (2) · Change Log 2026-09-10. Background and migration cautions are collected on the V4 Pro transition page.
Specifications (official feature table)
| Item | V4.1 Flash (deepseek-flash) |
|---|---|
| Model version | DeepSeek-V4.1-Flash |
| Context length | 1M tokens |
| Max output | 384K tokens (listed as MAXIMUM: 384K) |
| Thinking mode | Both non-thinking and thinking (thinking is the default) |
| Vision (image input) | Supported ✓ (the previous V4 Pro was not) |
| Tool calls / JSON output | Supported ✓ |
| Responses API / Anthropic API | Supported ✓ (both OpenAI-format and Anthropic-format endpoints) |
| Chat Prefix Completion (Beta) | Supported ✓ |
| FIM Completion (Beta) | Non-thinking mode only |
| Concurrency limit | 2500 (V4 Pro was 500) |
| Base URLs | https://api.deepseek.com (OpenAI format) / https://api.deepseek.com/anthropic (Anthropic format) |
Source: api-docs.deepseek.com Models & Pricing (retrieved 2026-09-11). See DeepSeek's Rate Limit & Isolation page for concurrency details.
Official benchmarks (published 2026-09-10)
Numbers as published by DeepSeek in its change log. Not independently re-measured by this site.
| Benchmark | Score |
|---|---|
| GPQA Diamond | 90.9 |
| HLE | 36.8 (39.1) — *tested on the pure-text subset only |
| Codeforces (Rating) | 3471 |
| MathArena Apex | 65.6 |
| Terminal-Bench 2.1 | 90.6 |
| Terminal-Bench 3.0 | 30.0 |
| Terminal-Bench 4.0 | 31.2 |
| DeepSWE v1.1 | 74.2 |
| NL2Repo-Bench | 65.4 |
| CyberGym | 88.1 |
| SEC-Bench Pro | 62.8 |
| ProgramBench | 20.3 |
| ExploitGym | 15.3 |
| HLE (w/ tools) | 63.9 |
| Automation-Bench | 54.8 |
| Agents' Last Exam | 31.8 |
| Chartography (w/ tools) | 78.9 |
| BabyVision (w/ tools) | 89.6 |
| ZeroBench-main (w/ tools) | 49.0 |
Source: Change Log "DeepSeek-V4.1-Flash Release" (2026-09-10). These are vendor-reported figures.
Setup in VS Code GitHub Copilot Chat (3 steps)
The steps the operator (okamo channel) actually used with the V4 family. Extension names and model-picker labels can change, so check the official docs too.
Get an API key
Sign up at platform.deepseek.com. Top up with PayPal or a credit card (VISA/Mastercard), then create an API key beginning with sk-. It is fully prepaid, with a $2 minimum top-up.
Install the extension
Install DeepSeek V4 for Copilot Chat (publisher: Vizards) from the VS Code Marketplace.
Configure and start
Run DeepSeek: Set API Key from the command palette (Cmd/Ctrl+Shift+P) to register your key, then pick a DeepSeek V4 model in the Copilot Chat model picker.
💡 If you set the model name in code, use deepseek-flash. Code that still points at deepseek-v4-pro will be answered by V4.1 Flash from 13:00 JST on 2026-09-14 (see the notes above).
Operator's setup record (2026-07-04, at the time of V4 Pro): okamo channel notes
FAQ
Q. How much does V4.1 Flash cost?
A. Per 1M tokens: input (cache miss) $0.15 off-peak / $0.30 peak, input (cache hit) $0.003 off-peak / $0.006 peak, output $0.60 off-peak / $1.20 peak. Peak hours are 01:00-04:00 and 06:00-10:00 UTC on weekdays only (JST 10:00-13:00 / 15:00-19:00); weekends are off-peak all day. Off-peak is half the peak rate.
Q. What happens if I keep calling deepseek-v4-pro after September 14?
A. From 13:00 JST on 2026-09-14 (12:00 Beijing time), and until V4.1 Pro is released, all requests to deepseek-v4-pro are routed to V4.1 Flash and billed at the V4.1 Flash price. Your code keeps working, but the model serving the request and the rate you pay are V4.1 Flash.
Q. When will DeepSeek V4.1 Pro be released?
A. No release date has been announced as of 2026-09-11. The official docs only say the routing lasts "until the future release of V4.1 Pro", with no committed date. This site will update as soon as an official announcement appears.
Q. Can it read images?
A. Yes. V4.1 Flash is built on the new architecture family with native multimodal visual understanding, and the official feature table marks Vision as supported (V4 Pro did not). For a cross-model view of image, audio, video and PDF support, see the multimodal LLM guide.
Q. Can I keep using deepseek-v4-flash?
A. Those names are still accepted for compatibility, but the underlying models are retired: requests are served by V4.1 Flash and billed at the V4.1 Flash price. New code should use deepseek-flash.
Q. Context length and max output?
A. 1M-token context and 384K-token maximum output. Thinking mode supports both non-thinking and thinking, with thinking as the default.
Sources
- DeepSeek API Docs — Models & Pricing (pricing, specs, notes; retrieved 2026-09-11)
- DeepSeek API Docs — Change Log (2026-09-10, DeepSeek-V4.1-Flash Release) (release details, benchmarks)
- DeepSeek official news (news260910)
- Related: pricing (DeepSeek) · token calculator · the V4 Pro transition · multimodal guide · release timeline
⚠️ Disclaimer
- Pricing, specs and benchmarks on this page are based on DeepSeek's official pages (api-docs.deepseek.com) as retrieved on 2026-09-11. Prices and model availability may change without notice. Always verify on the official page before contracting or production use.
- Benchmark figures are vendor-reported and have not been independently re-measured by this site.
- The setup steps reflect the operator's environment (recorded in July 2026 on V4 Pro) and may display differently after extension updates.
- This site is not a recommendation or an agent for any provider. DeepSeek is a trademark of DeepSeek. GitHub and GitHub Copilot are trademarks or registered trademarks of GitHub, Inc.