DeepSeek
DeepSeek introduces V4.1 Flash and new Flash pricing
DeepSeek says V4.1 Flash is cheaper and more capable than V4 Pro, with revised Flash series pricing taking effect on 10 September 2026.
DeepSeek
DeepSeek says V4.1 Flash is cheaper and more capable than V4 Pro, with revised Flash series pricing taking effect on 10 September 2026.
anthropic
Anthropic's economics team has published a tool that calculates how AI might affect growth and wages through 2030. Across three scenarios US output rises 1.6%, 8.3% or 32.4%. In the extreme case labour's share falls from 60% to 45.2%.
NVIDIA
NVIDIA announced CUDA Rust, two open-source projects that compile Rust GPU kernels natively and catch aliasing bugs at compile time.
OpenAI
OpenAI describes how an MIT researcher uses GPT-5.6 Sol with Codex to run quantum computing experiments autonomously, analyze results and calibrate qubits.
Artificial Intelligence
A joint US advisory names DeepSeek, Moonshot AI, Alibaba, MiniMax, StepFun and Z.AI, alleging they extracted billions of tokens from American frontier models since 2024.
OpenAI
OpenAI says an internal model solved the 90-year-old Navier-Stokes problem, but a mathematician who published a related result a day earlier is asking about Codex data.
anthropic
Jacob Coxon, who spent three years on pretraining research at OpenAI and then Anthropic, resigned and set out why. Anthropic's own alignment lead publicly agreed with him, putting the risk above 10% within the next decade.
anthropic
Evan Hubinger, who leads an Anthropic safety team, says the chance AI kills all humans within the decade is greater than 10 percent, hours after a colleague resigned over safety.
AWS
Students at 132 universities in 18 countries get a free year of Kiro, the same agent professional engineering teams use. That is 1,000 credits a month, verified by university email, with no credit card required. The March pilot covered 11 universities.
Physion Labs
Physion-Arc 1.0 Anime Track scored eight agents on 19 metrics. OpenArt leads with 77.8. Read the cost column and the table shifts: Invideo charges $50.7 per video for 69.2 points, while MiniMax delivers 68.5 for $6.3.
Cohere
Cohere has released under Apache-2.0 a serving system that fuses the entire decode step into one CUDA kernel. It is 1.58x faster than vLLM on a single H100, but it runs one model only, on H100 only, at batch sizes 1 to 8.
Eraser
Mermaid, D2 and Graphviz hand layout to an engine; the language cannot say where a box belongs. The format Eraser has opened lets the model emit coordinates directly. The reason: models learned coordinate layout in late 2025.