DeepSeek is shipping open models under permissive licenses and opening its own tooling: DeepSeek-V4.1-Flash-1017 under MIT, DeepInfer as an open inference stack, and dsh-superagent as a research prototype coding agent. It also runs a paid API whose peak rate is $0.3 per million input tokens and $1.2 per million output tokens for V4.1-Flash-1017, halved off-peak.
DeepSeek's recent coverage is a run of open releases. DeepSeek-V4-Flash-0731 arrived on August 7 with MIT-licensed weights and 284 billion total parameters, 13 billion active, and a 1 million token context. Two days later the same model shipped as DeepSeek-V4-Flash-0731 under a commercial-use license. By September 11 the line had moved to DeepSeek-V4.1-Flash, a mixture-of-experts model with 552 billion total parameters and image input, which the stories say beat Claude Opus 5 and GPT-5.6 Sol on multiple benchmark tests. On October 19 came DeepSeek-V4.1-Flash-1017 under the MIT license, positioned as faster and listed at $0.3 per million input tokens and $1.2 per million output tokens at peak, halved off-peak.
Price is the constant. V4 Pro 0813 shipped quietly on August 14 at 0.435 dollars per million input tokens and 0.87 per million output, described as roughly one fifty-seventh the cost of Claude Fable 5 with comparable performance. That followed an August 6 API price increase on V4 Flash, which DeepSeek attributed to unprecedented demand. Third-party testing complicated the value case: Composio ran 30 multi-step workflows across four harnesses and only 6 fully succeeded on V4 Flash.
The tooling has moved outward from the API. DeepSeek Harness 0.2 added Windows and macOS desktop apps on October 1, with plugins for periodic task execution and voice input. DeepInfer, the inference system behind DeepSeek's own services, was published on October 21 built on vLLM and running V4.1-Flash, without a KV cache system implemented; a companion deepseek-recipe converts between third-party API formats and the official DeepSeek format. On October 26, Strata developer Niko Veit said the quantized V4.1-Flash runs on a 64GB PC at 4 to 6 tokens per second on a GeForce RTX 5070. The newest release, dsh-superagent on November 2, is a research prototype that watches one GitHub repository, judges issues and pull requests, and can open its own pull requests.
Written with AI from the cited Yomimono stories below, each itself sourced.
See our editorial policy. Every fact links
the story it came from.
Nov 2
DeepSeek has released dsh-superagent, a research prototype coding agent built to watch a single GitHub repository. It monitors new issues, pull requests and developer group chats, judges what it finds, and can change code and open pull requests. Permissions and confirmation rules are configurable. DeepSeek says the code is not an official product and asks users to review it before use.
Oct 26
Niko Veit, developer of the open-source inference engine Strata, said on October 24 that the DeepSeek-V4.1-Flash model now runs on a PC with 64GB of main memory. The quantized release uses the GSQ-RCO compression method from DASLab at the Institute of Science and Technology Austria. Strata spreads model data across VRAM, main memory, and SSD. The 2-bit build measured 4 to 6 tokens per second on a GeForce RTX 5070 with 64GB of memory.
Oct 21
DeepSeek published DeepInfer, the inference system its own AI services use, with core functionality packaged for outside deployment. It is built on vLLM with ideas from SGLang and TensorRT LLM, runs DeepSeek-V4.1-Flash, and supports the DSpark acceleration technology, but a KV cache system is not implemented. DeepSeek also released deepseek-recipe, which converts requests and outputs between third-party APIs and the official DeepSeek API format.
Oct 19
DeepSeek released the open model DeepSeek-V4.1-Flash-1017 on October 17, 2026, positioned as an updated, faster version of DeepSeek-V4.1-Flash. On the DeepSWE v1.1 coding agent benchmark it scores above GPT-6 Astra, GPT-6.1 Sol and Claude Fable 5 while keeping cost low, and it recorded the highest scores on multiple tests. Peak pricing is $0.3 per million input tokens and $1.2 per million output tokens, halved off-peak.
Sep 11
DeepSeek released DeepSeek-V4.1-Flash on September 10, 2026. The open model is a mixture-of-experts system with 552 billion total parameters, 8 billion active parameters on input and 16 billion on output, and it supports image input. It beat Claude Opus 5 and GPT-5.6 Sol on multiple benchmark tests. Inference efficiency cut KV cache tokens to 1/3.9 of DeepSeek-V4-Flash, and off-peak API pricing is half the peak rate.
Aug 9
DeepSeek has released DeepSeek-V4-Flash-0731 as an open model under a commercial-use license. The MoE model has 284 billion total parameters and 13 billion active parameters. Third-party tests by Artificial Analysis rate its intelligence on par with Google's Gemini 3.6 Flash, and it outperforms GLM-5.2 in coding and GPT-5.6 Luna in agent performance.