DeepSeek V4 Pro vs V4.1 Flash-1017: price and specs

Quick answer: DeepSeek V4 Pro is still available, and it costs more than V4.1 Flash. DeepSeek dropped its plan to route V4 Pro requests to V4.1 Flash and now says it will keep serving V4 Pro with unchanged billing. At peak hours, deepseek-v4-pro (DeepSeek-V4-Pro-0813) costs $1.32 per million input tokens and $3.96 per million output tokens. The deepseek-flash model, which has served DeepSeek-V4.1-Flash-1017 since 17 October 2026, costs $0.30 and $1.20. Both have a 1M-token context window and up to 384K output tokens, but only Flash accepts images, and it allows 2,500 concurrent requests against 500 for V4 Pro. Off-peak prices are half the peak rates on both.

Key facts

• Model names: deepseek-flash serves DeepSeek-V4.1-Flash-1017, and deepseek-v4-pro serves DeepSeek-V4-Pro-0813.
• Peak price per million tokens: Flash $0.30 input, $0.006 cache hit, $1.20 output. V4 Pro $1.32 input, $0.044 cache hit, $3.96 output.
• Off-peak price, half the peak rate: Flash $0.15 / $0.003 / $0.60. V4 Pro $0.66 / $0.022 / $1.98.
• Both: 1M-token context, 384K max output, thinking mode by default plus a non-thinking mode, JSON output and tool calls.
• Flash only: image input. Concurrency limit: 2,500 requests for Flash, 500 for V4 Pro.
• Size, per DeepSeek's model card: V4.1 Flash has 552B backbone parameters, with 8B active per token in prefill and 16B in decode. V4 Pro has 1.6T, with 49B active.

Is DeepSeek V4 Pro still available?

Yes. When it released V4.1 Flash on 10 September, DeepSeek said it was phasing out V4 Pro: from 14 September, deepseek-v4-pro requests would route to V4.1 Flash at Flash prices until a V4.1-Pro launched. That release note still reads that way, but DeepSeek's changelog has the current plan:

DeepSeek adds that it will give notice if that changes, and it hasn't given a date for V4.1-Pro. Its pricing page lists both models side by side. The older names deepseek-v4-flash and deepseek-v4-flash-vision-exp still work for now, but their models are retired: DeepSeek serves those requests with V4.1-Flash-1017 at the Flash price.

Price: V4 Pro costs about 3 to 7 times as much

At the same hour, V4 Pro costs 4.4 times as much as Flash for uncached input, 7.3 times as much for cache hits and 3.3 times as much for output. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays. Every other hour, weekends included, is off-peak at half price.

• Example at peak: 1M uncached input tokens plus 100K output tokens costs $0.42 on Flash ($0.30 + $0.12) and $1.72 on V4 Pro ($1.32 + $0.40).
• The same job off-peak: $0.21 on Flash and $0.86 on V4 Pro.

Specs side by side

• Context window: 1M tokens on both. Max output: 384K tokens on both.
• Modes: both default to thinking and also offer a non-thinking mode. FIM completion (beta) works only in non-thinking mode.
• API features on both: JSON output, tool calls, the Responses API, the Anthropic-format API and chat prefix completion (beta).
• Vision: Flash takes images and text. V4 Pro is text-only.
• Concurrency limit: 2,500 for Flash, 500 for V4 Pro.
• Weights: V4.1-Flash-1017 is open under the MIT license on Hugging Face. The model card counts 552B backbone parameters plus a separate 196B-parameter Engram memory; Hugging Face's own metadata shows 763B parameters for the 1017 weights.

What's new in V4.1 Flash-1017

DeepSeek released V4.1-Flash-1017 on 17 October on its website, app and API, with open weights on Hugging Face. API calls and prices didn't change: deepseek-flash now points to the new version. DeepSeek says it scores higher, finishes tasks faster and writes shorter outputs than the previous V4.1 Flash. Its own numbers from the model card, vendor-reported:

• Terminal-Bench 4.0: 42.3, up from 31.2.
• DeepSWE v1.1: 77.0, up from 74.2.
• ProgramBench: 29.1, up from 20.3.

The 1017 card doesn't include V4 Pro. On the earlier V4.1 Flash card, DeepSeek's table put the previous Flash ahead of V4 Pro on every coding-agent test it lists, for example 74.2 against 62.7 on DeepSWE v1.1, while V4 Pro scored higher on the GPQA Diamond and HLE reasoning tests. Both cards are DeepSeek's own results, and they don't always use the same test harness. DeepSeek also recommends updating its DeepSeek Harness coding agent to v0.2.1-rc.1 or later for the best results with 1017.

Search & Fetch, also new on 17 October

The same day, DeepSeek opened its Search & Fetch API to everyone, powered by its own search engine. Each endpoint costs $0.75 per 1,000 successful requests. Calls made through server-side tools in DeepSeek Harness, Claude Code and Codex are free until 31 October 2026 at 16:00 UTC, which is midnight on 1 November in Beijing.