A project called llama-modchip, released on GitHub, claims to compress the 284.33B-parameter DeepSeek-V4-Flash mixture-of-experts model from roughly 77GB down to about 57-59GB of memory, letting it run on a 64GB CPU-only Linux machine without a GPU. The developers report prefill speeds near 44 tokens per second and CPU decode speeds of roughly 9-15 tokens per second, achieved through custom quantization, cache-tiling and pruning techniques, plus a separate mode running a 30B Nemotron model at much higher speeds on CPU hardware.
github.com
· 2026-11-08
Village's LLM Atlas, which tracks the strongest AI model at every price point, now shows open-weight models winning only 3% of budget tiers, down from 41% in June. Most of that decline followed OpenAI's GPT-6 Luna and GPT-6.1 Terra releases along with Anthropic's Claude Haiku 5.5, which reclaimed budgets previously held by DeepSeek V4 Flash and Zhipu's GLM-5.2.
village.dev
· 2026-11-06
A research team built a physics-based RL simulation of Tesla's flying Roadster, complete with wind gusts and the ramp-flip and Cybertruck kickflip stunts from the reveal, then had several leading LLMs try to pilot it. Models including DeepSeek V4.1 Flash, Nemotron 3.5 Super, Muse Spark Open, Beam 1.0, and GPT-6 Astra mostly crashed, landing few or none of the maneuvers even with unlimited thinking time, while a small 4B model fine-tuned with reinforcement learning landed reliably, including in a 33 m/s storm.
skypilot.ai
· 2026-10-30
DeepSeek released a point update to its open-weight V4.1 Flash model, a 552B-parameter Mixture-of-Experts system with 16B active parameters and a 1-million-token context window that handles both text and images. The company says the new revision needs only a quarter of the HBM and an eighth of the SSD storage for its KV cache compared with the prior version, using 890 bytes per token entry, and its internal benchmarks place the model between GPT-5.6 Sol and Claude Opus 5, beating the latter on some task tests.
tomshardware.com
· 2026-10-23
A test of 12 AI models from ten different labs asked each for its most controversial opinion of humans, three times per model, producing 35 answers. Researchers found 29 of those answers independently converged on a single idea—that people prioritize social belonging and group loyalty over objective truth—with phrasing varying but the core claim repeating across models like GPT-6.1 Sol, Grok 4.7, DeepSeek V4 Pro, Kimi K3 and Gemini 4 Argon.
modelcompare.ai
· 2026-10-22
DeepSeek has released DeepInfer, the inference framework it uses internally to serve its large-scale production systems, as an open-source project. The initial release supports the DeepSeek-V4.1-Flash model and includes optimized kernels, parallelism strategies, speculative decoding, asynchronous scheduling, and batch-invariant deterministic inference.
github.com
· 2026-10-20
DeepSeek has released V4.1 Flash 1017 (Max), a multimodal model accepting text and image input and producing text output, with a 1 million token context window. The model scored 45 on the Artificial Analysis Intelligence Index, well above the comparable median of 19, while running at 230 tokens per second.
artificialanalysis.ai
· 2026-10-19
DeepSeek has shipped V4.1 Flash 1017, dated October 17, 2026, as a snapshot of its V4.1 Flash model available via the deepseek-flash API. The sparse mixture-of-experts model uses a new Causal Encoder-Decoder architecture, activating 8B parameters on input and 16B on output from a 552B-parameter backbone, with native image understanding trained jointly with text from the start.
openrouter.ai
· 2026-10-17
DeepSeek has launched V4.1-Flash-1017, a minor version update to its V4.1-Flash model that the company says improves reasoning, coding, agentic task handling, and visual benchmark scores compared to the prior version. The update also shortens task execution time and output length. DeepSeek is also releasing deepseek-recipe, a Rust-based toolkit with Python bindings for encoding prompts and parsing model outputs.
huggingface.co
· 2026-10-17
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.
tomshardware.com
· 2026-10-09
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.
dwarfstar.sh
· 2026-10-02
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.
tomshardware.com
· 2026-10-01
GoKawiil