DeepSeek Releases V4.1-Flash-1017 With 77.0 DeepSWE Score, Opens Search APIs

DeepSeek said the checkpoint scores 77.0 on DeepSWE v1.1, up from 74.2 for the September V4.1-Flash, and beats the previous release on all 11 benchmarks it reported by 2.0 to 11.1 points. In DeepSeek's table, that DeepSWE score is above Claude Opus 5 at 74.0. Claude Opus 5 still leads on HLE without tools at 56.3 versus 42.6 for the new model, and on Terminal-Bench 4.0 at 51.8 versus 42.3.

The model uses 552 billion backbone parameters and 196 billion Engram parameters, activating 8 billion parameters per token in prefill and 16 billion in decode, the report said. It supports a 1 million-token context and up to 384,000 output tokens. Architecture details come from the V4.1-Flash technical report, which describes a multimodal mixture-of-experts model with 40 Transformer layers split into a 20-layer causal encoder and 20-layer decoder.

DeepSeek said V4.1-Flash-1017 uses reinforcement learning with self-feedback for end-to-end project work, with examples from game creation to 3D modeling. It is tuned for DeepSeek Harness, Tencent CodeBuddy and WorkBuddy, and DeepSeek recommends Harness v0.2.1-rc.1 or later. The company did not disclose hardware requirements or the 1017 weights; the September V4.1-Flash weights, at 763 billion parameters in FP8, are MIT-licensed on Hugging Face.

On pricing, developers can call the model as deepseek-flash through OpenAI- or Anthropic-format endpoints. Peak pricing is $0.30 per 1 million cache-miss input tokens and $1.20 per 1 million output tokens, with off-peak rates at half. The Search API returns ranked web results and the Fetch API returns a single page as Markdown; both cost $0.75 per 1,000 successful requests, with empty results and failures free, and each is limited to 5 queries per second per account.

DeepSeek said its in-house search engine serves hundreds of millions of requests daily. Search options include 1 to 50 results, optional page content and operators such as site:, lang:, after: and before:; results carry a 0-to-3 rerank score and a 0-to-4 authority tier. Server-side calls through DeepSeek Harness, Claude Code or Codex stay free until Nov. 1, 2026.

Editor's Summary

DeepSeek released V4.1-Flash-1017 with higher reported scores and opened Search and Fetch APIs, while Claude Opus 5 retains leads on some benchmarks. The release combines a lower-cost coding model with web tools for agent builders, though weights and hardware details for the new checkpoint were not disclosed.