Mistral released Large 4, a trillion-parameter mixture-of-experts model, in a research preview this week, with full weights due at the end of October. Independent benchmarking firm Artificial Analysis scored it 38 on its Intelligence Index, making it the top model from outside the US and China, though several Chinese open models including DeepSeek V4.1 Flash and MiMo-V2.6-Pro scored as high or higher overall. Large 4 scored 50 on Artificial Analysis's Cyber Index, tying GLM-5.3-Flash, and hit 82% on the CyberGym-E2E-AA cybersecurity test, its best individual result.
A developer reverse-engineered the Nokia 110 4G's firmware to turn its Calculator app into a chat interface that connects to DeepSeek's chat API. The prototype lets users type messages on the number pad and have the AI agent control phone functions like checking battery level, toggling the torch, starting calls, and setting alarms. The app is loaded into RAM from a computer and currently must be reloaded after every restart or power-off since permanent installation isn't yet solved.
A developer reports using DeepSeek 4.1 Flash heavily across a dozen projects for about a month, saying it performs comparably to models like Opus in daily work and conversation without a noticeable quality gap. The model is paired with a $10/month OpenCode Go subscription, making usage effectively unlimited, with typical sessions costing under $1.
DeepSeek, a Chinese AI company, is considering increasing its latest funding round from an initial 50 billion yuan to as much as 100 billion yuan ($14.9 billion). The company is attracting interest mainly from government-backed and corporate investors, including Tencent, CATL, and Geely Auto, while turning away private individual funds. The fundraising process is ongoing, with a target valuation of around 500 billion yuan ($75 billion).
Reflection AI has introduced Beam, a 501-billion-parameter open-weight model designed for reasoning, coding, and agentic tasks. Trained on nearly 24 trillion tokens, Beam claims to match Chinese models' performance while requiring significantly less inference compute. The company positions Beam as a cost-effective alternative to both Western and Chinese AI models, targeting enterprise and public sector applications.
Salvatore Sanfilippo, the creator of Redis, has released DwarfStar 4 (ds4), a lightweight C-based inference engine for running large open-weight language models locally on high-memory Mac, CUDA and ROCm hardware. The tool supports models including DeepSeek V4, V4.1 Flash, GLM 5.x and Qwen3.8 Flash Next, covering text and vision, with a local API, CLI and native agent, and is released under the MIT license.
DeepSeek has launched Harness, an open-source desktop application for macOS and Windows built on Cordis's plugin-based architecture. The tool can also be run as a web UI launched from code, and supports tasks like file organization, coding assistance, research, and background script execution.
DeepSeek, working with Huawei, has published open-source programming libraries for Huawei's Ascend AI chips, including DeepGEMM-Ascend for matrix computations and DeepEP-Ascend for chip-to-chip communication, plus Ascend support for the TileLang programming language. The tools were developed and tested on Ascend 950 hardware, including a 128-chip supernode system, according to a September 30 Reuters report.
The Call Center Doctors rented a four-GPU Nvidia H200 server on September 27 to run DeepSeek V4.1 Flash for its Claude Code coding agents instead of Claude Opus 5.5. The firm found the on-demand box cost $440.88 a day versus $184-$223 a day to get the same work done through DeepSeek's API, and it ultimately switched back to Opus 5.5, which worked out cheaper than keeping the rented hardware running.
DeepSeek has published a series of KV cache optimizations—including MLA architecture, Compressed Sparse Attention, and now CSA2 with cross-layer reuse and FP4 caching—that reportedly shrink cache memory footprint per token by roughly 437x compared to its earlier V1 model. These techniques reduce the VRAM needed to serve long-context sessions such as coding assistants, and the report says other labs have begun using similar approaches without crediting DeepSeek.
Rep. Ro Khanna, top Democrat on the House Select Committee on China, sent letters to Director of National Intelligence Jay Clayton and to DeepSeek, Alibaba, and Moonshot AI. He asked ODNI to assess US readiness for an AI system losing control, similar to an OpenAI agent incident involving Hugging Face, and to evaluate China's approach to catastrophic AI risk. He also asked the three Chinese companies for documentation on their pursuit of 'superintelligence,' recursive self-improvement, and any safety safeguards.
China has expanded its overseas travel restrictions on top AI and chip professionals to now include their spouses and children, according to people familiar with the matter. These relatives must now obtain approval from Beijing before traveling abroad, even for short trips, extending curbs previously applied only to executives and researchers at firms like Alibaba and DeepSeek since earlier this year.
GoKawiil