DeepSeek V4.1 Flash hands-on tests and comparisons

The author notes 1017 wanders less in agent loops, is already routed behind deepseek-flash, was optimized for DeepSeek Harness and Tencent CodeBuddy/WorkBuddy, and prices Search at ¥5 per 1,000 calls, while still splitting architect/challenger/reviewer roles to curb hallucination.

Official minor update: live on web, app and API with MIT-licensed weights on Hugging Face; DeepSeek reports 77 vs 74.2 on DeepSWE v1.1; Search and Fetch cost $0.75 per 1,000 calls, free via Claude Code, Codex and DeepSeek Harness until Oct 31.

Morph announces DeepSeek V4.1 1017 is available on OpenRouter and claims across-the-board improvements in agentic coding, without specific benchmarks or task numbers.

ModelScope hosts the 1017 model, described as delivering stronger coding and agent performance with faster task completion and shorter outputs.

The author spent four days comparing Opus 5.5, GPT 6.1 Sol, Gemini 4 Argon and DeepSeek V4.1 Flash on the same greenfield app workflow.

On the same task Opus 5.5 finished in 20 minutes for $6 while V4.1 Flash took a full hour for just $0.23, roughly 26x cheaper, with Opus winning but DeepSeek close behind.

A breakdown of the release: MIT open weights, shorter outputs cutting agent cost and latency, 77%+ on DeepSWE at about $0.20 per task versus $0.50 for the prior Flash, plus Search/Fetch APIs at $0.75 per 1k calls.

Experiential Labs lists DeepSeek V4.1 Flash 1017 as free for a promotion with limits, offering ~1.05M context and up to 393K output tokens.

Using V4.1 Flash 1017 and the latest DeepSeek Harness with no skills, the author produced an Agentic RL Lab promo video from a single prompt for under ¥3 and 40M tokens.

On the same voxel 3D reconstruction of a Tesla Roadster, DeepSeek V4.1 Flash 1017 cost $0.89 versus $1.07 for GPT-6 Luna; Flash runs locally or via API.

Povilas Korop says the first benchmark results for DeepSeek V4.1 Flash 1017 are astonishing and he is re-running tests to confirm, with a video promised tomorrow.

A user says Flash-1017 is already strong and wonders whether the upcoming V4.1 Pro could challenge Gemini 4 Astra.