DeepSeek V4.1-Flash Cuts Agent Memory Eightfold
DeepSeek V4.1-Flash drops to 890 bytes of cache per token under an MIT licence, with flat API pricing and a one million token context.
Articles tagged open weight
Posted inAI News
GLM-5.3-Flash Runs on 100,000 Chinese Chips
GLM-5.3-Flash was the stealth model Ox Alpha: Zhipu confirms it, ships MIT weights and reveals a cluster of 100,000 Chinese chips.
Posted inAI News
Qwen3.8-Flash-Next Cuts AI Prices Twelvefold
Qwen3.8-Flash-Next lands twelve times cheaper than Qwen3.8-Max, with benchmarks aimed at DeepSeek and Claude Opus 4.6.
Posted inAI News
Qwen3.8-27B Is Now Free to Download and Modify
Alibaba released Qwen3.8-27B under Apache 2.0: 27 billion parameters, 262,000 tokens of native context, already live on Hugging Face and ModelScope.
Posted inOur AI Tool Tests
Kimi K3 Test Ranks It First on Frontend Code
Kimi K3 test: first on the frontend arena and level with Opus 4.8 on simple code. Where it drops off, and which workloads it actually fits.
DeepSeek V4 Flash Nears GPT-5.6 Luna at 60% Lower Cost
DeepSeek V4 Flash "0731" scores 50 on the Intelligence Index, one point behind GPT-5.6 Luna, in MIT open weights and 60% cheaper.
Posted inAI News
Anthropic’s Open-Weights Position, Explained
Dario Amodei says Anthropic never called for an open-weights ban and lays out three targeted measures instead, from chip controls to mandatory safety tests.
Posted inAI News
Kimi K3 Ships Its Open Weights, 1.4TB Free
Moonshot ships the open weights of Kimi K3, 2.8 trillion parameters at 1.4TB to download for free, and the model leads a major coding benchmark.
Posted inAI News
DeepSeek V4 Retires Its Old Model Names July 24
DeepSeek V4 becomes the only entry point: the old deepseek-chat and deepseek-reasoner names stop answering on July 24 at 15:59 UTC.
Posted inAI News
Moonshot Halts Kimi Sales as Its GPUs Max Out
Moonshot suspended new Kimi K3 subscriptions after compute capacity maxed out in forty-eight hours. Existing subscribers keep their access.