DEEPSEEK · MODEL RELEASE TRACKER
Deepseek-V4.1-Flash-1017
Deepseek-V4.1-Flash-1017 is a released update to Deepseek's Flash family, published with open weights on Hugging Face and ModelScope. It was trained with reinforcement learning using self-feedback, is tuned for software-development and agentic tasks, and the company reports improved benchmark performance and lower per-task costs versus its predecessor and some larger rivals.CURRENT SNAPSHOT2/5 DIMENSIONS WITH DATA
The dimensions that change the decision.
Cost / Efficiency on DeepSWEOn the DeepSWE benchmark, V4.1-Flash-1017 reached a top score of 77.0% at an average cost of just under $0.20 per task; in a lower-cost setting it achieved about 72% at roughly $0.10 per task.
Not established from the available sources.
Not established from the available sources.
Benchmarks / ComparisonsIn Deepseek's internal benchmarks, V4.1-Flash-1017 outperforms its predecessor V4.1-Flash and is reported to beat larger competing models (including Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3) on several coding and AI agent categories.
Not established from the available sources.
RELEASE TIMELINE
Published, source-backed release events only.
VERIFIABLE FACTS
Every value stays attached to a source and date.
RELEASE · Availability / ReleaseDEVELOPER CLAIM
Deepseek-V4.1-Flash-1017 was released and is available through Deepseek's website, app, and API. The model weights were published as open source on Hugging Face and ModelScope.
CAPABILITIES · Training method and capabilitiesDEVELOPER CLAIM
The model was trained/updated using reinforcement learning with self-feedback (RLSF). Deepseek reports the model runs faster, produces shorter outputs, and performs well on end-to-end software development and agent tasks; it can also animate compact explainer videos. It was tuned for development environments such as Deepseek Harness (v0.2.1-rc.1 or later recommended), Tencent CodeBuddy, and WorkBuddy.
BENCHMARKS · Benchmarks / ComparisonsDEVELOPER CLAIM
In Deepseek's internal benchmarks, V4.1-Flash-1017 outperforms its predecessor V4.1-Flash and is reported to beat larger competing models (including Claude Opus 5, GPT 5.6-Sol, Kimi K3, and GLM 5.3) on several coding and AI agent categories.
PRICING · Cost / Efficiency on DeepSWEDEVELOPER CLAIM
On the DeepSWE benchmark, V4.1-Flash-1017 reached a top score of 77.0% at an average cost of just under $0.20 per task; in a lower-cost setting it achieved about 72% at roughly $0.10 per task.
WHAT CHANGED