DeepSeek-V4-Flash Update

Date: 2026-07-31​

DeepSeek-V4-Flash Update​

The charismatic merchandise of the DeepSeek-V4-Flash API is now successful nationalist beta. The API calling method remains unchanged — simply group the exemplary sanction to deepseek-v4-flash to usage the latest version.

Significantly enhanced supplier capabilities, pinch benchmark results acold exceeding V4-Pro-Preview:

• Terminal Bench 2.1: 82.7
• NL2Repo: 54.2
• Cybergym: 76.7
• DeepSWE: 54.4
• Toolathlon verified: 70.3
• Agent Last Exam: 25.2
• Automation Bench (Public): 25.1
• DSBench-FullStack: 68.7
• DSBench-Hard: 59.6

Note 1: For the Code Agent tasks successful the nationalist benchmark sets, the charismatic DeepSeek-V4-Flash was tested utilizing the DeepSeek Harness minimal mode (to beryllium released soon) arsenic the framework, pinch the max effort level, topp=0.95, and temperature=1.0
Note 2: DSBench-FullStack is an soul full-stack improvement trial set, and DSBench-Hard is an soul Coding Agent hard-problem trial set

The charismatic V4-Flash natively supports the Responses API format and is specifically adapted for Codex. For the circumstantial configuration, please mention to the documentation.

DeepSeek-V4-Flash-0731 keeps the aforesaid exemplary architecture and size arsenic DeepSeek-V4-Flash-preview, and was only re-post-trained.

Note: This update only upgrades the DeepSeek-V4-Flash API. The DeepSeek-V4-Pro API and the APP/WEB models are unchanged.
The charismatic merchandise of DeepSeek-V4-Pro will travel soon.

Date: 2026-04-24​

DeepSeek-V4​

The DeepSeek API now supports V4-Pro and V4-Flash, disposable via some the OpenAI ChatCompletions interface and the Anthropic interface. To entree the caller models, the base_url remains unchanged, and the exemplary parameter should beryllium group to deepseek-v4-pro aliases deepseek-v4-flash.

The 2 bequest API exemplary names, deepseek-chat and deepseek-reasoner, will beryllium discontinued successful 3 months (2026-07-24). During the existent period, these 2 exemplary names constituent to the non-thinking mode and reasoning mode of deepseek-v4-flash, respectively.

For much details, please mention to this documentation.

Date: 2025-12-01​

DeepSeek-V3.2​

Both deepseek-chat and deepseek-reasoner person been upgraded to DeepSeek-V3.2.

• deepseek-chat corresponds to DeepSeek-V3.2's non-thinking mode
• deepseek-reasoner corresponds to DeepSeek-V3.2's thinking mode

DeepSeek-V3.2-Speciale​

DeepSeek-V3.2-Speciale is served via a impermanent endpoint: base_url="https://api.deepseek.com/v3.2_speciale_expires_on_20251215". Same pricing arsenic V3.2, nary instrumentality calls, disposable until Dec 15th, 2025, 15:59 (UTC Time).

For much details, please mention to this documentation.

Date: 2025-09-29​

DeepSeek-V3.2-Exp​

Both deepseek-chat and deepseek-reasoner person been upgraded to DeepSeek-V3.2-Exp.

• deepseek-chat corresponds to DeepSeek-V3.2-Exp's non-thinking mode
• deepseek-reasoner corresponds to DeepSeek-V3.2-Exp's thinking mode

For much details, please mention to this documentation.

Date: 2025-09-22​

DeepSeek-V3.1-Terminus​

Both deepseek-chat and deepseek-reasoner person been upgraded to DeepSeek-V3.1-Terminus. deepseek-chat corresponds to DeepSeek-V3.1-Terminus's non-thinking mode, while deepseek-reasoner corresponds to its thinking mode.

This update maintains the model's original capabilities while addressing issues reported by users, including:

• Language consistency: Reduced occurrences of Chinese-English mixing and occasional abnormal characters;
• Agent capabilities: Further optimized the capacity of the Code Agent and Search Agent.

Date: 2025-08-21​

DeepSeek-V3.1​

Both deepseek-chat and deepseek-reasoner person been upgraded to DeepSeek-V3.1. deepseek-chat corresponds to DeepSeek-V3.1's non-thinking mode, while deepseek-reasoner corresponds to its thinking mode.

• Key updates successful DeepSeek-V3.1:
  • Hybrid reasoning architecture: A azygous exemplary supports some reasoning mode and non-thinking mode
  • Improved reasoning efficiency: Compared to DeepSeek-R1-0528, DeepSeek-V3.1-Think provides answers successful importantly little time
  • Enhanced supplier capabilities: With post-training optimization, the caller exemplary achieves awesome improvements successful instrumentality usage and intelligent supplier tasks
    • SWE-bench Verified: 66.0
    • SWE-bench Multilingual: 54.5
    • Terminal-bench: 31.3

Date: 2025-05-28​

deepseek-reasoner​

deepseek-reasoner Model Upgraded to DeepSeek-R1-0528:

• Enhanced Reasoning Capabilities
  • Significant benchmark improvements (Pass@1)
    • AIME 2025: 70.0 → 87.5 (+17.5)
    • GPQA: 71.5 → 81.0 (+9.5)
    • LCB_v6: 63.5 → 73.3 (+9.8)
    • Aider: 57.0 → 71.6 (+14.6)
  • Note: Complex reasoning tasks whitethorn devour much tokens compared to bequest R1 version.
• Optimized Front-end Development
  • Generated web pages and games now characteristic improved aesthetics.
• Reduced Hallucinations
  • Significantly suppressed mirage issues coming successful bequest R1 version.
• JSON Output & Function Calling Support
  • Function telephone performance:
    • Tau-bench score: 53.5 (Airline) / 63.9 (Retail)

Date: 2025-03-24​

deepseek-chat​

deepseek-chat Model Upgraded to DeepSeek-V3-0324:

• Enhanced Reasoning Capabilities

  • Significant improvements successful benchmark performance:
    • MMLU-Pro: 75.9 → 81.2 (+5.3)
    • GPQA: 59.1 → 68.4 (+9.3)
    • AIME: 39.6 → 59.4 (+19.8)
    • LiveCodeBench: 39.2 → 49.2 (+10.0)

• Optimized Front-End Web Development

  • Improved accuracy successful codification generation
  • More aesthetically pleasing web pages and crippled front-ends

• Upgraded Chinese Writing Proficiency

  • Enhanced style and contented quality:
    • Aligned pinch the R1 penning style
    • Better value successful medium-to-long-form writing

• Feature Enhancements

  • Improved multi-turn interactive rewriting
  • Optimized translator value and missive writing

• Improved Chinese Search Capabilities

  • Enhanced study study requests pinch much elaborate outputs

• Function Calling Improvements

  • Increased accuracy successful Function Calling, fixing issues from erstwhile V3 versions

Date: 2025-01-20​

deepseek-reasoner​

• deepseek-reasoner is our caller exemplary DeepSeek-R1. You tin invoke DeepSeek-V3 by specifying model='deepseek-reasoner'.
• For details, please mention to: DeepSeek-R1 Release
• For guides, please mention to: Thinking Mode

Date: 2024-12-26​

deepseek-chat​

• The deepseek-chat exemplary has been upgraded to DeepSeek-V3. The API remains unchanged. You tin invoke DeepSeek-V3 by specifying model='deepseek-chat'.
• For details, please mention to: introducing DeepSeek-V3

Date: 2024-12-10​

deepseek-chat​

The deepseek-chat exemplary has been upgraded to DeepSeek-V2.5-1210, pinch improvements crossed various capabilities. Relevant benchmarking results include:

• Mathematical: Performance connected the MATH-500 benchmark has improved from 74.8% to 82.8% .
• Coding: Accuracy connected the LiveCodebench (08.01 - 12.01) benchmark has accrued from 29.2% to 34.38% .
• Writing and Reasoning: Corresponding improvements person been observed successful soul trial datasets.

Additionally, the caller type of the exemplary has optimized the personification acquisition for record upload and webpage summarization functionalities.

Date: 2024-09-05​

deepseek-coder & deepseek-chat Upgraded to DeepSeek V2.5 Model​

The DeepSeek V2 Chat and DeepSeek Coder V2 models person been merged and upgraded into the caller model, DeepSeek V2.5.

For backward compatibility, API users tin entree the caller exemplary done either deepseek-coder aliases deepseek-chat.

The caller exemplary importantly surpasses the erstwhile versions successful some wide capabilities and codification abilities.

The caller exemplary amended aligns pinch quality preferences and has been optimized successful various areas specified arsenic penning tasks and instruction following:

• ArenaHard triumph complaint improved from 68.3% to 76.3%
• AlpacaEval 2.0 LC triumph complaint accrued from 46.61% to 50.52%
• MT-Bench people roseate from 8.84 to 9.02
• AlignBench people accrued from 7.88 to 8.04

The caller exemplary has further enhanced its codification procreation capabilities based connected the original Coder model, optimized for communal programming exertion scenarios, and achieved the pursuing results connected the modular trial set:

• HumanEval: 89%
• LiveCodeBench (January-September): 41%

Date: 2024-08-02​

API Launches Context Caching connected Disk Technology​

The DeepSeek API has innovatively adopted difficult disk caching, reducing prices by different bid of magnitude.

For much specifications connected the update, please mention to the archiving Context Caching is Available 2024/08/02.

Date: 2024-07-25​

New API Features​

• Update API /chat/completions
  • JSON Mode
  • Function Calling
  • Chat Prefix Completion（Beta）
  • 8K max_tokens（Beta）
• New API /completions
  • FIM Completion（Beta）

For much details, please cheque the archiving New API Features 2024/07/25

Date: 2024-07-24​

deepseek-coder​

The deepseek-coder exemplary has been upgraded to DeepSeek-Coder-V2-0724.

Date: 2024-06-28​

deepseek-chat​

The deepseek-chat exemplary has been upgraded to DeepSeek-V2-0628.

Model's reasoning capabilities person improved, arsenic shown successful applicable benchmarks:

• Coding: HumanEval Pass@1 79.88% -> 84.76%
• Mathematics: MATH ACC@1 55.02% -> 71.02%
• Reasoning: BBH 78.56% -> 83.40%

In the Arena-Hard evaluation, the triumph complaint against GPT-4-0314 accrued from 41.6% to 68.3%.

The model's role-playing capabilities person importantly enhanced, allowing it to enactment arsenic different characters arsenic requested during conversations.

Date: 2024-06-14​

deepseek-coder​

The deepseek-coder exemplary has been upgraded to DeepSeek-Coder-V2-0614, importantly enhancing its coding capabilities. It has reached the level of GPT-4-Turbo-0409 successful codification generation, codification understanding, codification debugging, and codification completion. Additionally, it possesses fantabulous mathematical and reasoning abilities, and its wide capabilities are connected par pinch DeepSeek-V2-0517.

Date: 2024-05-17​

deepseek-chat​

The deepseek-chat exemplary has been upgraded to DeepSeek-V2-0517. The exemplary has seen a important betterment successful pursuing instructions, pinch the IFEval Benchmark Prompt-Level accuracy jumping from 63.9% to 77.6%. Additionally, connected API end, we person optimized exemplary expertise to travel instruction filled successful the ``system" part. This optimization has importantly elevated the personification acquisition crossed a assortment of tasks, including immersive translation, Retrieval-Augmented Generation (RAG), and more.

The model's accuracy successful outputting JSON format has been enhanced. In our soul trial set, the JSON parsing complaint accrued from 78% to 85%. By introducing due regular expressions, the JSON parsing complaint was further improved to 97%.