Search “DeepSeek”

English translation is being prepared.

On October 20, 2026, DeepSeek announced the open-sourcing of its internal production-grade inference framework, DeepInfer. According to Latent Space, the framework powers DeepSeek-V4.1-Flash, is customized based on vLLM, supports distributed inference, speculative decoding, SWA Bounded Replay, and structured output, and targets the NVIDIA SM100 and Huawei Ascend 950 platforms; the published single-cluster production figures are 155 trillion tokens processed per day, a 96% input cache hit rate, 400 ms first-token latency, and a generation speed of 230 tokens/s. The report also noted that components such as KV cache management and load balancing have not yet been fully open-sourced. The next day, DeepSeek published the DeepInfer technical report on Zhihu, disclosing internal details including the hardware selection for Prefill and Decode and the KVCache targets, and providing data showing a single cluster serving 155 trillion Tokens/day at $38 per card per hour at off-peak prices. The framework is currently open-sourced, but the KV cache system, Prefill/Decode disaggregation, and load balancing required for a complete service are not fully included in this open-source project.

View story:DeepSeek open-sources inference framework DeepInfer

DeepSeek open-sources inference framework DeepInfer

On October 20, 2026, DeepSeek open-sourced its internal inference framework DeepInfer. According to Yuchen Jin, citing data disclosed by DeepSeek, a single cluster processes 155 trillion tokens per day and generates $38 in revenue per GPU-hour (off-peak pricing); DeepInfer supports both NVIDIA Blackwell and Huawei Ascend. A subsequent 36Kr report stated that its accompanying open-source deepseek-recipe supports NVIDIA SM100 GPUs and Huawei Ascend 950 NPUs. On October 21, swyx, citing Latent.Space data, said a single cluster processes 155T tokens per day with a 96% input cache hit rate, and argued that the lifetime value of each sold GPU has risen by about 30%. On October 24, Nathan Lambert said DeepSeek's single cluster processes 150T tokens per day, exceeding the combined open-source model inference volume of OpenRouter, Fireworks, Together, and BaseTen, and noted that the actual number of clusters is unknown; on the same day, LMSYS published a same-hardware benchmark comparison, stating that SGLang meets or exceeds DeepInfer on DeepInfer's own benchmark, with prefill throughput up to 1.04x and decode step length shortened by 1.1%–7.4% on 64x GB300.

View story:DeepSeek open-sources inference framework DeepInfer

English translation is being prepared.

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

Arena announces DeepSeek-V4.1-Flash-1017 tops the Agent Arena open-source leaderboard

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

English translation is being prepared.

On October 20, 2026, Foreign Ministry spokesperson Mao Ning, responding at a regular press conference to reports that the US government has asked its agencies to stop using artificial intelligence tools from Chinese companies such as DeepSeek, said that China consistently opposes overgeneralizing the concept of national security and politicizing science, technology, and economic and trade issues, and will resolutely safeguard the legitimate rights and interests of Chinese companies. On the same day, reports said that the US Department of Homeland Security had ordered the removal of DeepSeek-related technology from most government systems and barred civilian government agencies from using DeepSeek and any technology built with its models; the reports said the ban took effect at the end of July and that agencies under the Department of Defense are not covered. Other reports said the order, based on the 2018 Federal Acquisition Supply Chain Security Act, requires civilian agencies to remove DeepSeek and related products from their information technology systems, as well as third-party products and services developed using DeepSeek, with exceptions requiring an application submitted 60 days in advance with reasons stated.

View story:Foreign Ministry responds to US demand to stop using DeepSeek

DeepSeek completes nearly 100 billion yuan in financing

DeepSeek has completed a new round of financing, raising close to 100 billion yuan. According to a Huxiu report, this round raised a minimum of 80 billion yuan and could ultimately approach 100 billion yuan, with the pre-money valuation maintained at 500 billion yuan; founder Liang Wenfeng personally contributed about 20 billion yuan. The report also noted that in the first seven months of 2026 the company's revenue was 475 million yuan and its net loss was 715 million yuan, with the API business gross margin at 82.9%. The funds are mainly directed to a GW-level intelligent computing center in Ulanqab, Inner Mongolia, where the company plans to deploy 160,000 Huawei Ascend 950DT chips; the company plans to submit its prospectus by the end of 2026 and list on the STAR Market in 2027. No subsequent factual developments have been disclosed so far.

View story:DeepSeek Completes Nearly 100 Billion Yuan Financing

DeepSeek-V4.1-Flash-20261017 (Max) Tops Arena Open Model Leaderboard

Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has entered Arena and ranks first among open models on both the Agent Arena and Code Arena: WebDev leaderboards. The news was released by Arena on October 21, 2026, and is the latest verifiable development to date. Arena also noted that the 3D world generated by the model and more detailed ranking information can be viewed in the related post. Apart from this top-ranking result, there are no other subsequent developments with qualified factual evidence.

DeepSeek-V4.1-Flash-20261017 (Max) Tops the Arena Open Model Leaderboard

Arena announced that DeepSeek-V4.1-Flash-20261017 (Max) has entered Arena and ranks first among open models on both the Agent Arena and Code Arena: WebDev leaderboards. The news was released by Arena on October 21, 2026, and is the latest verifiable development to date. Arena also noted that the 3D world generated by the model and more detailed ranking information can be viewed in the related post. Apart from this top-ranking result, there are no other subsequent developments with qualified factual evidence.

View story:DeepSeek-V4.1-Flash-20261017 (Max) Tops Arena Open Model Leaderboard

Unsloth releases GGUF for running DeepSeek V4.1-Flash-1017 locally

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

Author uses Codex and DeepSeek-V4.1-Flash-1017 to generate a four-quadrant sticky note app

On October 25, 2026, an author used the Chinese-localized version of Codex, DeepSeek-V4.1-Flash-1017, and the Hermes agent to generate a four-quadrant sticky note app from a single prompt, and automatically uploaded and published it to GitHub. According to the same report, the finished product supports Chinese and English, a time capsule, free resizing of sticky notes, and local saving, and is packaged as a single-file portable app; the first version generated a static index.html in about six to seven minutes, and subsequent iterations changed it to a sticky note format, with the whole process costing a little over four yuan. There are currently no subsequent independent development facts.

The author used Codex and DeepSeek-V4.1-Flash-1017 to generate a four-quadrant sticky note app

On October 25, 2026, an author used the Chinese-localized version of Codex, DeepSeek-V4.1-Flash-1017, and the Hermes agent to generate a four-quadrant sticky note app from a single prompt, and automatically uploaded and published it to GitHub. According to the same report, the finished product supports Chinese and English, a time capsule, free resizing of sticky notes, and local saving, and is packaged as a single-file portable app; the first version generated a static index.html in about six to seven minutes, and subsequent iterations changed it to a sticky note format, with the whole process costing a little over four yuan. There are currently no subsequent independent development facts.

View story:Author uses Codex and DeepSeek-V4.1-Flash-1017 to generate a four-quadrant sticky note app

WorkBuddy announces limited-time free access to DeepSeek-V4.1-Flash-1017 international version

WorkBuddy announced that its international version has been updated to DeepSeek-V4.1-Flash-1017 and is available free of charge for a limited time until October 31. A report published on October 17 stated that the WorkBuddy international version had integrated this version; the cited original DeepSeek post said that this version performs better on benchmarks, completes tasks faster, and produces shorter outputs, and that it has launched on the official website, App, and API, with optimizations for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy, and others, and recommended upgrading to DeepSeek Harness v0.2.1-rc.1 or higher. A report on October 18 further clarified that this free-access offer applies only to the international version, and that the domestic version has not yet launched a corresponding free-access promotion. There are currently no subsequent developments.

DeepSeek releases DeepSeek-V4.1-Flash-1017

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

WorkBuddy announces limited-time free access to DeepSeek-V4.1-Flash-1017 international edition

WorkBuddy announced that its international version has been updated to DeepSeek-V4.1-Flash-1017 and is available free of charge for a limited time until October 31. A report published on October 17 stated that the WorkBuddy international version had integrated this version; the cited original DeepSeek post said that this version performs better on benchmarks, completes tasks faster, and produces shorter outputs, and that it has launched on the official website, App, and API, with optimizations for DeepSeek Harness, Tencent CodeBuddy, WorkBuddy, and others, and recommended upgrading to DeepSeek Harness v0.2.1-rc.1 or higher. A report on October 18 further clarified that this free-access offer applies only to the international version, and that the domestic version has not yet launched a corresponding free-access promotion. There are currently no subsequent developments.

View story:WorkBuddy announces limited-time free access to DeepSeek-V4.1-Flash-1017 international version

DeepSeek releases DeepSeek-V4.1-Flash-1017

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

English translation is being prepared.

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

Arena announces DeepSeek-V4.1-Flash-1017 as the top open-source model on the Code Arena leaderboard

On October 17, 2026, DeepSeek released DeepSeek-V4.1-Flash-1017, which the company says delivers stronger benchmark performance, faster task completion, and shorter outputs, achieving a Pareto improvement in performance and cost; it is now available on the official website, App, and API, and the model also launched on OpenRouter the same day. The company also announced that the Search and Fetch APIs are open to everyone, priced at $0.75 per 1,000 calls per endpoint, with calls made through server-side tools in DeepSeek Harness, Claude Code, and Codex temporarily free, and billing starting at 16:00 UTC on October 31, 2026. On October 17, Thomas Wolf commented that DeepSeek called this a "minor update," yet Terminal-Bench 4.0 rose from 31.2 to 42.3. On October 19, Arena announced that the model, with a score of 1670, became the highest-ranked open-source model on Code Arena: WebDev. On October 20, OpenCode updated it to version 1017 in Go and Free; Unsloth released a GGUF that can run locally, with the 1-bit quantized version retaining about 88.2% accuracy and shrinking in size from 508GB to 250GB; Arena then announced that it topped the Agent Arena open-source model leaderboard.

View story:DeepSeek releases DeepSeek-V4.1-Flash-1017

DeepSeek Completes Nearly 100 Billion Yuan Financing

DeepSeek has completed a new round of financing, raising close to 100 billion yuan. According to a Huxiu report, this round raised a minimum of 80 billion yuan and could ultimately approach 100 billion yuan, with the pre-money valuation maintained at 500 billion yuan; founder Liang Wenfeng personally contributed about 20 billion yuan. The report also noted that in the first seven months of 2026 the company's revenue was 475 million yuan and its net loss was 715 million yuan, with the API business gross margin at 82.9%. The funds are mainly directed to a GW-level intelligent computing center in Ulanqab, Inner Mongolia, where the company plans to deploy 160,000 Huawei Ascend 950DT chips; the company plans to submit its prospectus by the end of 2026 and list on the STAR Market in 2027. No subsequent factual developments have been disclosed so far.

English translation is being prepared.

English translation is being prepared.

View story:English translation is being prepared.