DeepSeek just open-sourced dsh-superagent-prototype, an experimental background agent designed to handle GitHub tasks without cluttering human attention. It runs tasks in isolated sessions and only pings developers when a pull request is ready or human intervention is strictly required. Keep in mind that DeepSeek frames this as research code rather than an official product.
ISTA-DASLab just released non-uniform GSQ-RCO GGUF quantizations for DeepSeek-V4.1-Flash-1017, compressing the 763B MoE down to as low as 1.75 bpw. Paired with the Strata inference engine, the massive model runs on a single consumer 12GB GPU while the 2.00 bpw build retains 100% of the base model's task performance. Why it matters: Running frontier-tier open models locally no longer requires a server rack.
DeepSeek just released and open-sourced DeepSeek-V4.1-Flash-1017, a coding model trained with self-feedback reinforcement learning to tackle end-to-end projects from 3D modeling to full games. It promises faster task completion with tighter outputs across web, app, and API. DeepSeek also opened its proprietary Search and Fetch APIs, which remain free for API users through October 31, 2026.
DeepSeek just open-sourced DeepInfer, the production inference engine behind its DeepSeek-V4.1-Flash model. Crucially, the framework supports Huawei Ascend 950 NPUs alongside Nvidia SM100 GPUs, letting teams deploy without Nvidia hardware. Why it matters: It offers a practical path toward diversifying inference silicon, though production serving still requires an external KV cache and load balancing.
Anthropic and OpenAI are privately war-gaming scenarios for public and political fallout following a catastrophic AI disaster, like a massive cyberattack on critical infrastructure. Industry insiders warn a major event could happen in the next 6 to 12 months. Companies are racing to educate Congress to soften inevitable regulatory crackdowns before disaster strikes.
DeepSeek just launched DeepSeek-V4.1-Flash, a 552B-parameter MoE model with native visual understanding. Built on a Causal Encoder-Decoder architecture, it activates just 8B parameters on input and 16B on output, outpacing the flagship V4-Pro while slashing KV cache to 1/4 the HBM and 1/8 the SSD storage.
DeepSeek just rolled out DeepSeek-V4-Flash-Vision-Exp on its API platform. The experimental multimodal model retains V4-Flash's text and reasoning chops while closing in on Opus-4.8 on multimodal agent benchmarks. Images cap out at 384 tokens each, and a new Files API lets developers upload once and reuse across calls.
DeepSeek just launched DeepSeek-V4-Pro into general availability, introducing major agent upgrades alongside flexible reasoning levels (low, high, and max) across V4-Pro and V4-Flash. It natively supports OpenAI's Responses API for Codex and rolls out off-peak API pricing that cuts token costs by 50%.
DeepSeek just open-sourced DeepSeek Harness (dsh), an agent framework built entirely on an "everything-is-a-plugin" architecture. Powered by Cordis, it turns models, tools, and sandboxes into modular extensions and spins up an out-of-the-box local Web UI. It is currently in active developer preview, so expect breaking changes ahead.
DeepSeek just officially released DeepSeek-V4-Flash-0731 with built-in DSpark speculative decoding. Thanks to heavy post-training, this lightweight model leapfrogs the larger V4-Pro preview across terminal and coding benchmarks, punching well into Opus 4.8 territory while fully open-sourced under an MIT license.
Chinese authorities just held talks with leading tech firms to discuss potentially restricting overseas access to the country's most advanced AI models, including DeepSeek and unreleased systems, according to Reuters sources. The restrictions are being weighed amid escalating geopolitical tensions. Why it matters: If enacted, the policy would lock out global developers and international businesses that have come to rely on Chinese frontier AI.
DeepSeek just launched and open-sourced DeepSeek-V4 Preview, making a 1M token context standard across its services. It delivers two MoE models—the flagship V4-Pro and the lightweight V4-Flash—built around hybrid compressed attention that slashes KV cache memory to just 2% of standard architectures. It basically tackles the persistent context blowout in long-running coding agents, and the new models are already live in chat and via API.
Sakana AI just launched Sakana Chat, powered by its new Namazu alpha series that adapts open-weight frontier LLMs for Japan through post-training. The models slash DeepSeek's 72% question refusal rate on sensitive topics to nearly zero while keeping benchmark parity in reasoning and coding. The web chat also packs real-time search capabilities.
A senior Trump administration official claims Chinese startup DeepSeek trained its upcoming model on banned Nvidia Blackwell chips, reportedly clustered in an Inner Mongolia data center. Officials also suspect the firm will scrub technical markers of chip usage. The leak further inflames Washington's bitter debate over export control enforcement.
Anthropic just accused three Chinese AI labs—DeepSeek, Moonshot, and MiniMax—of running industrial-scale distillation attacks. The firms allegedly routed over 16 million exchanges through roughly 24,000 fraudulent accounts to siphon Claude's reasoning and coding powers. Why it matters: Anthropic warns that distilled copies strip out safety guardrails and directly undermine US export controls.
OpenAI just warned US lawmakers that Chinese startup DeepSeek is bypassing access restrictions to distill US models like ChatGPT for its own training. In a congressional memo, OpenAI claimed DeepSeek employees used obfuscated routers and automated scripts to harvest outputs while cutting corners on safety. DeepSeek and parent firm High-Flyer have not commented.
ByteDance and Alibaba are reportedly prepping next-generation flagship models for release during the February Lunar New Year holiday. ByteDance is lining up Doubao 2.0 alongside major image and video upgrades, while Alibaba is readying Qwen 3.5 with targeted reasoning boosts. With DeepSeek V4 also rumored for mid-February, China's frontier AI race is entering an intense new clash.
DeepSeek just launched DeepSeek-V3.2 and DeepSeek-V3.2-Speciale, two reasoning-first models built specifically for AI agents. The standard V3.2 is now live across its app, web interface, and API, while the higher-end Speciale reasoning model is currently limited to the API.
Deepseek just open-sourced an OCR system that compresses image-based text by up to 10x while retaining 97% of information. By drastically reducing vision tokens to as few as 64 per page, it lets language models process massive context without hitting compute limits.
DeepSeek just released an experimental model called V3.2-exp that uses a sparse attention method to cut long-context API costs in half. The system pairs a lightning indexer with a fine-grained token selector to ease server load over long prompts. It is already open-weight on Hugging Face, opening the door for third-party benchmarking.
