Learn about this first
DeepSeek has upgraded V4.1 Flash to DeepSeek-V4.1-Flash-1017, with simultaneous updates to the web version, app, and API. The API calling method and pricing remain unchanged, and it is still accessed via deepseek-flash; the model weights remain available for download under the MIT license. The official announcement says the new version delivers stronger benchmark performance, faster task completion, and shorter outputs, representing a Pareto improvement in performance and cost, and recommends upgrading to DeepSeek Harness 0.2.1-rc.1 or later.
At the same time, DeepSeek has for the first time opened the search engine behind its app to developers, adding two interfaces, Search and Fetch, priced at ¥5 (approximately $0.75) per 1,000 calls. Server-side calls through DeepSeek Harness, Claude Code, and Codex are free until November 1, 2026.
Third-party evaluations show strong performance: Arena says the model ranks first among open-source models in both Agent Arena and Code Arena: WebDev, and 11th overall in Agent Arena; Artificial Analysis gives it an Intelligence Index score of 45, tied for second among open-weight models; and it scores 78.1% on ARC-AGI-2. ARC Prize additionally says that in the maximum reasoning mode, the model used 36% fewer reasoning tokens for its first action than the original version (the average across 25 public games), and 48.6% fewer across 55 semi-private games.
AI-generated from reports · updated 14 days agoLatest developments
ARC Prize Publishes DeepSeek V4.1 Flash Test Results
Event progress15
ARC Prize Publishes DeepSeek V4.1 Flash Reasoning Token Efficiency Data
ARC Prize: ARC Prize Publishes DeepSeek V4.1 Flash 1017's ARC-AGI (Verified) ResultsUnsloth Publishes DeepSeek-V4.1-Flash Quantized Model Evaluation Results
Unsloth: Unsloth Releases DeepSeek-V4.1-Flash GGUFUnsloth Releases DeepSeek-V4.1-Flash-1017 GGUF
Unsloth: Unsloth Releases the DeepSeek-V4.1-Flash-1017 Dynamic V3.1 GGUF Quantized VersionArtificial Analysis announces new Cyber Index rankings
Artificial Analysis: DeepSeek V4.1 Flash 1017 Tops the Artificial Analysis Cyber Index Open-Model RankingsSiliconFlow compares the 3D worlds generated by DeepSeek V4.1 Flash and GPT-6 Astra
SiliconFlow: DeepSeek V4.1 Flash vs. GPT-6 Astra Builds a 3D WorldARC Prize announces DeepSeek V4.1 Flash model results
ARC Prize: ARC Prize Publishes DeepSeek V4.1 Flash 1017's Results on ARC-AGI-1 and ARC-AGI-2DeepSeek-V4.1-Flash takes first place on both Arena leaderboards
Arena: DeepSeek-V4.1-Flash-20261017 (Max) Tops Arena's Agent and Code Leaderboards as the No. 1 Open ModelArena announces that DeepSeek-V4.1-Flash ranks first on the Agent Arena
Arena: DeepSeek-V4.1-Flash-20261017 (Max) Tops Agent Arena as the No. 1 Open ModelDeepSeek V4.1 Flash updated to 1017 in OpenCode Go/Free
opencode: OpenCode Updates DeepSeek V4.1 FlashArtificial Analysis publishes evaluation results for DeepSeek V4.1 Flash 1017
Artificial Analysis: DeepSeek V4.1 Flash 1017 Released, Scoring 45 on the Artificial Analysis Intelligence Index and Tying for Second Among Open ModelsSiliconFlow launches DeepSeek-V4.1-Flash-1017
SiliconFlow: DeepSeek-V4.1-Flash-1017 Launches on SiliconFlowArena announces that DeepSeek-V4.1-Flash tops the Code Arena
Arena: DeepSeek-V4.1-Flash-20261017 Tops the Code Arena: WebDev Open-Model RankingsWorkBuddy Intl updated to DeepSeek-V4.1-Flash-1017
Tencent WorkBuddy: WorkBuddy Intl updated to support DeepSeek-V4.1-Flash-1017, free until October 31OpenRouter launches the DeepSeek V4.1 Flash model
OpenRouter: DeepSeek V4.1 Flash 1017 launched on OpenRouterDeepSeek-V4.1-Flash-1017 update
DeepSeek: API changelog: DeepSeek-V4.1-Flash-1017 released, with improved results across multiple benchmarks and no change to API pricing