by DeepSeek

DeepSeek-V4-Pro review, pricing and verdict

DeepSeek's flagship open-weight reasoning model: a 1.6T-parameter MoE with 1M context, an MIT license and weekday peak and off-peak API rates.

Best for
Agentic coding
Not for
Multimodal or vision applications
  • Available
  • Open source

Should you use DeepSeek-V4-Pro?

Pick V4-Pro when you want coding and agent performance from weights you can download and run privately, or hosted rates that halve outside weekday peak hours. Skip it if you need image or audio input. Artificial Analysis, scoring it 36 on its Intelligence Index, still calls its per-token price expensive for an open-weight model of this size.

Best for

  • Agentic coding Preview-card scores of 80.6% on SWE-bench Verified and 93.5 on LiveCodeBench, with documented integrations for coding agents.
  • Long-document analysis A 1M-token context and 384K output cap handle full codebases or document sets in one request.
Show 2 more
  • Scheduled batch inference Off-peak rates are half the peak rates, so work that can wait for nights and weekends costs less.
  • Private or fine-tuned deployments MIT-licensed open weights allow fine-tuning, air-gapped deployment and self-hosted inference.

Not for

  • Multimodal or vision applications Pro is text-only; use DeepSeek-V4-Flash, Gemini 3.1 Pro or Claude Opus 4.8 for images.
  • Voice assistants No audio input or output, and max-effort reasoning adds long waits before the first answer token.
  • Highest-concurrency workloads on the hosted API The concurrency limit for Pro is 500, a fifth of the limit DeepSeek lists for Flash.

DeepSeek-V4-Pro pricing

Input, per 1M tokens
$1.32/1M
Output, per 1M tokens
$3.96/1M
Cached input, per 1M tokens
$0.04/1M
Pricing model
per-token
Free tier
5 million tokens for new accounts on DeepSeek API

What a real job costs

JobInputOutputTotal
Summarise a 20-page PDF$0.040$0.0040$0.044
Support reply$0.0026$0.0012$0.0038
One coding agent run$0.264$0.079$0.343

Budgets: 20-page PDF = 30k in / 1k out · Support reply = 2k in / 300 out · Coding agent run = 200k in / 20k out. Computed from the vendor's per-token prices at render time; cached-input discounts are not applied.

Cost examples

Process a 100K-token research paper and generate a 100K-token summary (peak rates)
$0.53
Daily agentic coding loop (1M tokens in, 200K out, peak rates)
$2.11
Customer-support deployment (1,000 turns at 2K in / 500 out per turn, peak rates)
$4.62

Prices are per 1M tokens and depend on the time of day. Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday, excluding Chinese public holidays: $1.32 cache-miss input, $0.044 cache-hit input and $3.96 output. Every other hour, including weekends and Chinese public holidays, is off-peak at half those rates: $0.66, $0.022 and $1.98. This schedule took effect at 16:00 UTC on August 16, 2026, replacing the earlier flat $1.74 and $3.48 list price and the 75% launch promotion that ended May 31, 2026. DeepSeek says prices may change. Artificial Analysis lists the peak rate. For comparison, the records for Claude Opus 4.8 list $5 and $25, GPT-5.5 $5 and $30, and Qwen3.7-Max $2.50 and $7.50 per 1M input and output tokens. Self-hosting the MIT weights carries infrastructure cost only.

Self-hosting requirements

Min vram gb
865
Self-hosted
Yes
Recommended GPU
4x H200 141GB multi-node cluster; or Fireworks/Lightning AI for managed high-speed inference

What DeepSeek-V4-Pro can do

DeepSeek-V4-Pro is DeepSeek's flagship open-weight Mixture-of-Experts model: 1.6 trillion parameters, 49 billion active per token, a 1M-token context window and an MIT license. It reached general availability on August 13, 2026, after an April preview. It takes text input only and offers low, high and max reasoning effort through the API.

What goes in and what comes out

Reads

  • text
  • tool-calls

Writes

  • text
  • tool-calls

Capabilities

  • Function callingSupported
  • Structured outputSupported
  • Tool useSupported
  • Audio inputNot supported
  • Audio outputNot supported
  • Code executionNot supported
  • Video inputNot supported
  • VisionNot supported
  • Web browsingNot supported
Reasoning modes
non-think, think-high, think-max
Long context recall
high

Key features

  • Reasoning Effort Control

    Thinking and non-thinking modes, with low, high and max effort levels added at the general release.

  • 1 Million Token Context Window

    Fits a full codebase or document set in one request, with the output cap set at 384K tokens.

  • MIT License Open Weights

    Full weights on Hugging Face at deepseek-ai/DeepSeek-V4-Pro with no restrictions on commercial use or fine-tuning.

  • OpenAI and Anthropic API Formats

    One API serves OpenAI-format requests, the Responses API and an Anthropic-format endpoint, with JSON output and tool calls.

  • Peak and Off-Peak Pricing

    Weekday peak windows of seven hours UTC carry double the rate of every other hour, weekends and Chinese public holidays.

Known limitations

  • Text-only input: no image, audio or video understanding on Pro.
  • Hosted concurrency is capped at 500 requests for Pro.

Who it is for

  • ML engineers running cost-sensitive agentic coding pipelines needing frontier benchmark scores at open-model pricing
  • Research teams requiring fine-tunable open weights with MIT license for private or domain-specific model adaptation
  • Enterprise teams processing large codebases or documents exceeding 128K tokens in a single context
  • Organizations evaluating open-weight alternatives to GPT-5.5 or Claude Opus 4.7 for cost reduction

Supported languages

  • English
  • Chinese
  • Spanish
  • French
  • German
  • Japanese
  • Korean
  • Arabic
  • Portuguese
  • Russian

DeepSeek-V4-Pro benchmarks

DeepSeek-V4-Pro against its nearest rivals

SWE-bench Verified % solved, higher is better

GPQA Diamond % correct, higher is better

Output speed tokens per second, higher is better

Where it sits

Priced around the middle of the 106 GA models with a published price (rank 55), in the top third on SWE-bench Verified (rank 12 of 39), and one of 90 whose vendor states it does not train on customer data.

MeasureThis modelRankMedian
Blended price (3:1)$ per 1M tokens$1.98/M#55 / 106peer median $1.73/M
Output speedtokens/s92 tok/s#31 / 59peer median 94 tok/s
SWE-bench Verified% solved80.6%#12 / 39peer median 76.5%
GPQA Diamond% correct90.1%#22 / 63peer median 87.8%

Ranks are against GA models on Hokai that publish the same figure; ties share a rank.

Benchmark scores

Show all 9 benchmark scores
  • LiveBench vendor-reported · 27 Apr 202693.5%A rolling set of fresh questions that cannot have been in training data, % correct.
  • MATH vendor-reported · 27 Apr 202692%Competition maths problems, % solved.
  • GPQA Diamond vendor-reported · 27 Apr 202690.1%PhD-level science questions that are hard to search for, % correct.
  • HumanEval vendor-reported · 27 Apr 202690%Small programs that must pass hidden tests, % passing.
  • MMLU vendor-reported · 27 Apr 202689%General-knowledge exam across 57 subjects, % correct.
  • MMLU-Pro vendor-reported · 27 Apr 202687.5%A harder version of the 57-subject knowledge exam, % correct.
  • SWE-bench Verified vendor-reported · 27 Apr 202680.6%Real GitHub issues fixed end to end, % solved.
  • Humanity's Last Exam vendor-reported · 27 Apr 202637.7%Expert-written questions across many fields, % correct.
  • AA Intelligence Index independent · 23 Oct 202636Composite of the evaluations Artificial Analysis runs for its index, 0 to 100.

Measured figures

AA blended price
$1.98/M
Output speed
92 tok/s

A benchmark is an exam, not the job. Scores transfer unevenly between tasks, so weigh the one closest to your workload and read every figure with its source.

Safety benchmarks

Jailbreak resistance
moderate

Strengths and limits

Strengths

  • The MIT-licensed weights can be downloaded, fine-tuned and run privately with no vendor dependency.
  • Long inputs stay in a single call, so a whole codebase or document set needs no chunking.
  • Off-peak API rates are half the peak rates, so batch-style and overnight workloads can be scheduled into the cheaper hours.
  • Scored 80.6% on SWE-bench Verified in Think Max mode on the April preview card.
  • MIT-licensed open weights with a 1M-token context and 384K-token output cap.

Limits

  • Text-only: DeepSeek's docs list vision as unsupported on Pro, so image work needs V4-Flash or another vendor's model.
  • Artificial Analysis rates the per-token price as expensive against open-weight models of similar size and calls the model somewhat verbose.
  • Self-hosting needs roughly 865GB of weights and a multi-GPU cluster, and the hosted API caps Pro at 500 concurrent requests.

DeepSeek-V4-Pro alternatives

ModelLabContextIn / out per 1MReads and writes
DeepSeek-V4-Pro This pageDeepSeek1M$1.32 / $3.96text → text
Qwen3.7-MaxPick DeepSeek V4 Pro for MIT-licensed downloadable weights; pick Qwen3.7-Max when its higher listed GPQA Diamond score (92.4 vs 90.1) matters more than owning the weights.Alibaba Cloud1M$2.50 / $7.50text → text
Claude Opus 4.8Pick DeepSeek V4 Pro for open weights and lower API rates; pick Claude Opus 4.8 for image and PDF input and its higher SWE-bench Verified score.Anthropic1M$5 / $25text, image, pdf → text
Hy4 flashTencent's Hy4 flash, released 16 November 2026, is a lower-cost multimodal model that reads text and images across a 1M-token window.Tencent1M$0.07 / $0.281text, image → text

DeepSeek-V4-Pro specs

Context window
1.0M tokens
Max output
384K tokens
Latency (P50)
2100 ms
Tokens per second
91.6 tok/s
Mixture of experts
Yes
Parameters
1.6T total, 49B active per token
Architecture
Mixture-of-Experts (MoE) with Hybrid Attention (CSA + HCA)
VRAM requirements
865GB at FP4+FP8 mixed precision; requires multi-GPU cluster (4x H200 141GB minimum)

Training data

Pre-trained on 33 trillion tokens with emphasis on long documents, code, math, scientific text, and agentic execution traces. Post-training uses two stages: SFT plus GRPO domain cultivation, then unified model consolidation via on-policy distillation. Sensitive personal data screened per April 2026 model card. Trained on Huawei Ascend 950PR chips.

Weight formats

  • FP4+FP8 mixed precision (official)
  • BF16
  • GGUF Q4_K_M
  • GGUF Q2_K

API endpoints

NameEndpointAuth
DeepSeek APIhttps://api.deepseek.com/v1/chat/completionsAPI Key
AWS Bedrockhttps://aws.amazon.com/bedrock/AWS IAM
Google Vertex AIhttps://cloud.google.com/vertex-aiGCP IAM
Azure AI Foundryhttps://azure.microsoft.com/en-us/products/ai-foundryAzure AD
Fireworks AIhttps://fireworks.ai/models/deepseek-ai/deepseek-v4-proAPI Key
Together.aihttps://www.together.ai/API Key
DeepInfrahttps://deepinfra.com/API Key
OpenRouterhttps://openrouter.ai/deepseek/deepseek-v4-proAPI Key

SDKs

  • Python
  • TypeScript
  • JavaScript
  • Go
  • Java
  • C#
  • Ruby

Regions

  • us-east-1
  • us-west-2
  • eu-central-1
  • ap-northeast-1
  • ap-southeast-1

Prompting patterns

Show 2 prompting patterns
  • Set reasoning mode explicitly per request

    Include reasoning_effort: 'high' or 'max' in the API call for complex tasks; omit for simple lookups to save tokens.

    Think Max mode costs more context budget tokens; Non-think is 3-5x faster for routine tasks.

  • Use OpenAI-compatible SDK with base_url swap

    client = OpenAI(base_url='https://api.deepseek.com/v1', api_key=DS_KEY); model='deepseek-v4-pro'

    DeepSeek API is fully OpenAI ChatCompletions-compatible; no SDK changes needed beyond base_url and model name.

Known quirks

  • Think Max requires 384K token context budget

    Workaround: Set context window to at least 384K tokens when using Think Max mode, otherwise the model falls back to shorter reasoning chains.

  • Rates double during weekday peak windows

    Workaround: Schedule non-urgent batch jobs outside 01:00-04:00 and 06:00-10:00 UTC on weekdays, or on weekends and Chinese public holidays, to pay the off-peak rate.

DeepSeek-V4-Pro facts

Trains on your data
No
Licence
MIT License · Open source
Safety posture
Permissive
Released
24 Apr 2026
Current version
deepseek-v4-pro
Family
DeepSeek-V4
Variant
Chat
Deployment
API · AWS Bedrock · Google Vertex AI · Azure · Self-hosted · Together · Fireworks · DeepInfra
Data retention
DeepSeek does not publish a specific API input retention window. Sensitive data handling is not covered by a public DPA on the direct API. Enterprise data governance requires routing through AWS Bedrock or Azure AI Foundry for contractual data handling guarantees.

Content restrictions

  • Safety alignment covers standard harmful content categories but is not extensively documented in a public safety card.
  • MIT license allows weight modification: downloaded models can be used without the original safety training in place.

Safety controls

Rlhf
Yes
Red teaming
not disclosed
System prompt
Yes
Constitutional AI
No
Content filter API
No
Moderation endpoint
No

Data governance

Iso 27001
No
Soc2 type2
No
HIPAA eligible
No
Trains on inputs
No
Data residency options
us, eu
Eu AI act classification
General-purpose AI (GPAI) with likely systemic risk obligations given 1.6T parameter scale

Version history

  • DeepSeek-V4-Pro-0813 · 13 Aug 2026

    General availability. Adds low, high and max reasoning effort and native OpenAI Responses API support. Peak and off-peak API pricing took effect 2026-08-16 16:00 UTC.

  • deepseek-v4-pro · 24 Apr 2026

    Initial preview release. SWE-bench 80.6% (Think Max), LiveCodeBench 93.5, Codeforces 3206. MIT license, 1M context, 384K max output. Three reasoning modes.

  • deepseek-v3.2 · 01 Jan 2026

    Prior generation. SWE-bench 67.8%, LiveCodeBench 74.1, 128K context. Predecessor to V4 series.

What changed

  • Pricing change 75% Promotional Price Cut Active Until May 31, 2026 source
  • Deprecation Legacy deepseek-chat and deepseek-reasoner Aliases Retire July 24, 2026 source
  • Release DeepSeek-V4-Pro Preview Launch: 1.6T MoE, 1M Context, MIT License source

Questions about DeepSeek-V4-Pro

How much does the DeepSeek-V4-Pro API cost per million tokens?

During the weekday peak windows it costs $1.32 per 1M cache-miss input tokens, $0.044 for cache-hit input and $3.96 for output. All other hours are off-peak at half that: $0.66, $0.022 and $1.98. The schedule started on August 16, 2026. Self-hosting the open weights has no per-token fee beyond GPU costs.

Can you try DeepSeek-V4-Pro for free?

New API accounts receive 5 million free tokens to start, with no separate free tier beyond that credit. Teams that want ongoing free access can instead self-host the MIT-licensed open weights, though that requires a multi-GPU cluster rather than a hosted plan.

Which models are alternatives to DeepSeek-V4-Pro?

Qwen3.7-Max is a hosted rival with a 1M context window and a higher listed GPQA Diamond score, but proprietary weights. Claude Opus 4.8 and the GPT flagship add image input and list higher SWE-bench Verified scores at higher per-token prices. DeepSeek-V4-Flash is the same family with image input.

How does DeepSeek-V4-Pro compare with Claude Opus 4.8?

Claude Opus 4.8 lists 88.6% on SWE-bench Verified and 93.6% on GPQA Diamond against V4-Pro's preview-card 80.6% and 90.1%, and it accepts images and PDFs where V4-Pro takes text only. V4-Pro counters with MIT-licensed downloadable weights and API rates well below the Opus 4.8 record's listed prices.

How do I start calling DeepSeek-V4-Pro?

Create an API key on the DeepSeek platform and point an OpenAI-compatible SDK at https://api.deepseek.com with the model name deepseek-v4-pro. An Anthropic-format endpoint is available at https://api.deepseek.com/anthropic. Choose thinking or non-thinking mode, and set reasoning effort to low, high or max. Self-hosters pull the weights from Hugging Face instead.

About DeepSeek-V4-Pro

DeepSeek-V4-Pro is the flagship model from DeepSeek, the Chinese AI lab founded in 2023 by High-Flyer hedge fund CEO Liang Wenfeng. It first appeared as a preview on April 24, 2026 under the MIT license, and DeepSeek announced general availability on August 13, 2026 with a new reasoning-effort control and native OpenAI Responses API support. The current API build is labelled DeepSeek-V4-Pro-0813. It is a Mixture-of-Experts transformer with 1.6 trillion total parameters and 49 billion active per token. According to the April model card it was pre-trained on 33 trillion tokens and trained on Huawei Ascend 950PR chips rather than Nvidia hardware. Its smaller sibling, DeepSeek-V4-Flash, trades capability for throughput and, unlike Pro, accepts images.

Read the full background

The context window is 1,048,576 tokens with up to 384,000 tokens of output per request, per DeepSeek's API docs. The API accepts OpenAI-format and Anthropic-format requests, and supports JSON output, tool calls, the Responses API, chat prefix completion (beta) and FIM completion in non-thinking mode (beta). Thinking and non-thinking modes are both available, and DeepSeek recommends low reasoning effort for simple tasks, high for daily agent workflows and max for complex ones. Pro does not support vision: input is text only, so image, audio and video work belongs on a sibling or a rival such as Gemini 3.1 Pro. Coding agents such as OpenCode and Claude Code are among the agent integrations DeepSeek documents for its API.

The benchmark scores stored on this page come from DeepSeek's April 27, 2026 model card for the preview checkpoint: 80.6% on SWE-bench Verified in Think Max mode, 90.1% on GPQA Diamond, 37.7% on Humanity's Last Exam, 93.5 on LiveCodeBench and a Codeforces rating of 3,206. DeepSeek published a new benchmark table for the 0813 build as an image on its announcement page, and those figures have not been transcribed here, so treat the stored scores as the preview's. Artificial Analysis, an independent tester, scores the 0813 Max configuration at 36 on its Intelligence Index (v4.3.2) with output speed of about 92 tokens per second, and says the model generated 160 million tokens running the index, which it calls somewhat verbose.

Teams running agentic coding pipelines, long-document analysis or private inference have the clearest use for it. The MIT license allows fine-tuning, air-gapped deployment and redistribution, which no closed rival such as Claude Opus 4.8 or GPT-5.5 offers. The other side of the ledger is real. Both of those rivals accept images and list higher SWE-bench Verified scores on their records, and Qwen3.7-Max lists a higher GPQA Diamond score. Artificial Analysis also rates V4-Pro's per-token price as expensive next to other open-weight models of similar size, so the case for it rests on the open weights and the off-peak rates rather than on being the cheapest model available. The concurrency limit on DeepSeek's own API is 500 for Pro, and self-hosting needs roughly 865GB of FP4 and FP8 weights and a multi-GPU cluster, with 4x H200 141GB the practical minimum cited in the April card's deployment notes.

DeepSeek's published safety documentation is thinner than that of Western frontier labs. The April model card says sensitive personal data and identification information were excluded from training data, and it discloses no external red-team partners. The MIT license also means anyone can download the weights and strip the safety tuning, a different risk profile from an API-only model. The version history on this page lists the preview, the 0813 general-availability build and its predecessor DeepSeek-V3.2, whose 128K context and 67.8% SWE-bench Verified score the V4 series replaced.

Sources and guides

More AI Models on Hokai

  • Nemotron-Labs-SigmaNVIDIA's Nemotron-Labs-Sigma, released in November 2026 in 8B and 3B sizes, writes text by denoising whole blocks of tokens in parallel, not one at a time.
  • Qwen4 MaxAlibaba's Qwen4 Max, released 16 November 2026, is a sparse mixture-of-experts model with a 1M-token context and text, image and video input.
  • Gemini 4 HydrogenGoogle DeepMind's November 2026 Gemini 4 model for coding agents, with four thinking levels from low to xhigh and a free tier in AI Studio.
  • Qwen4-Max-PreviewAlibaba's November 2026 Qwen 4 flagship preview, open only to Token Plan subscribers, with three reasoning levels from low to xhigh.
  • Claude Fable 5Superseded by Fable 5.1 and Fable 5.2 but still sold on the Claude API: Anthropic's first generally available Mythos-class model, released June 2026, with a 1M-token context.
  • Claude Fable 5.1Superseded by Claude Fable 5.2 in November 2026 but still sold on the Claude API: Anthropic's second Mythos-class model, released Sept 1, 2026, for long-horizon coding and research.