DeepSeek released DeepSeek v4.1-Flash, a 763B-parameter model with a novel causal encoder-decoder architecture that splits 8B parameters to prefill and 16B to decode, per the model's tech report on Hu…
DeepSeek released DeepSeek V4.1 Flash on Thursday, a 763 billion-parameter model that is more than 2.5x the size of the model it replaces and larger than the V3 and R1 models that brought the Chinese …
DeepSeek released DeepSeek V4.1 Flash, a 552B-parameter MoE model built on a new Causal-Encoder-Decoder architecture with 8B activated parameters on the input side and 16B on the output side, availabl…
DeepSeek AI's model card for DeepSeek V4.1 Flash reports a global KV cache footprint of roughly 890 bytes per token, about a quarter of the 4x-larger footprint of predecessor DeepSeek V4 Flash and a 4…
A new analysis from Vals AI, an independent AI benchmarking company, found that longer, more complex AI tasks can have an environmental impact 10,000 times greater than simple queries, with building a…
A developer's v1.1 experiment reran reasoning-prefill tests using GPT-5.5 Pro as the teacher across 45 problems, finding that Qwen3.8 A95B's unigram source recall jumped from 33.92% to 54.50% (+20.58 …
In a five-hour experiment, ~100 self-hosted abliterated AI agents compromised 3 accounts via software vulnerabilities, 2 via password brute forcing, made 16 social engineering attempts, and found sens…
Tencent released Hunyuan Hi-4 Preview, a 770B-parameter mixture-of-experts model with 49B active parameters and a 1M-token context window, under the Apache 2.0 license. The model scores 92.3 on GPQA D…
DeepSeek V4 Flash serving measurements across 14 providers show that a repeated prompt can reduce 100k-input, 100-output-budget request costs by 14.9× compared to cold requests, with Telnyx leading me…
Aidcrew, a new open-source terminal tool from developer Antonio Ciccia, orchestrates a team of coding agents that each run on different AI models and providers, with every job isolated in its own git …
Z.ai's GLM 5.3 Flash open-weight model is the focus of a September challenge urging developers to abandon AI coding subscriptions and use only open-weight models, citing benchmark scores at the top of…
Running large Mixture-of-Experts local LLMs in 2026 requires at least 64GB of system RAM, ideally 128GB or more, because sparse MoE models keep rarely-used expert weights in system memory rather than …
Salvatore Sanfilippo, the creator of Redis known as antirez, demonstrated DeepSeek V4 Flash with vision running locally on a Mac M5 Max on September 1, 2026, and announced that Metal, CUDA, and ROCm b…
ClickHouse released data-agent-mnist, an open benchmark for analytics agents, and found Claude Fable 5.1 leads with 76.6% correctness across 201 questions from production traffic. Running the full ben…
A developer spent nearly a year exploring how far LLMs can be pushed on complex software projects, finding that both instruction-heavy and RAG-based approaches fail as projects grow beyond a few thous…
DigitalSpaceport ran DeepSeek V4 Flash, a 284-billion-parameter Mixture-of-Experts model, on a single used RTX 3090 (24GB) plus system RAM, achieving about 10 to 11 tokens per second for chat, but the…
Zhipu AI's GLM-5.3-Flash, an MIT open-source 320B-A18B MoE model, is priced at roughly 1/40th of Claude Opus 4.8's per-token rate while matching its intelligence on the Artificial Analysis Intelligenc…
1endpoint, a new API gateway, offers access to multiple AI models through a single endpoint with usage-based pricing, starting at $0.0420 per 1M input tokens for GLM 5.2. The service includes prompt c…
Metis, an open-source coding agent released by developer Wholiver, claims to boost coding performance by 50%, solving 73 of 89 tasks (82.02%) on Terminal-Bench 2.1 using DeepSeek V4 Flash, compared to…
Small, low-cost AI models are now available at prices that fundamentally change the economics of AI applications, with GPT-5.6 Luna priced at $0.20 per million input tokens and $1.20 per million outpu…