Models
Frontier and open models: what shipped, what it can do, and how the claims hold up.
-
GPT‑6.1 Sol is cheaper than Astra, not than the Sol it replaces
Standard prices are unchanged from GPT‑6 Sol. Only cached input was halved. Every “one-fifth the cost” claim is measured against Astra.
-
H Company releases Holo4 open-weight computer-use agents
The Paris start-up’s open-weight agents work through screens, code and APIs, and it publishes every benchmark run behind its scores.
-
AMD to buy Fei-Fei Li’s World Labs in $8.2bn all-stock deal
AMD will pay about $8.2bn in stock for the spatial-intelligence lab, and Fei-Fei Li will become its chief scientist.
-
Anthropic releases Claude Sonnet 5.5, a faster, cheaper mid-tier model
Anthropic’s new mid-tier model keeps Sonnet 5’s prices but, the company says, runs more than 30% faster and costs up to 30%…
-
Grok 4.7 competes on price, not on top scores
SpaceXAI holds Grok 4.6 prices for its new flagship, which trails Fable 5.1 on most of the benchmarks it published. The pitch…
-
Google’s long-form video stack, scored on benchmarks it built itself
Four frameworks wrap Gemini and Veo to keep minutes-long video coherent, and the headline 81.4 comes from a benchmark the same team…
-
OpenAI rebuilds prompt caching for GPT-6
Cheaper reuse of repeated context, a 30-minute cache window, and two pages from the same publisher that state the discount differently.
-
OpenAI adds GPT-6 Sol and Luna, halving API prices
OpenAI’s two cheaper GPT-6 models undercut their predecessors by half, and the benchmark claims come with caveats worth reading.
-
NVIDIA documents its open speech-to-speech model, weak spots included
The technical report for NVIDIA’s open full-duplex speech model reports strong interruption handling alongside noticeably weaker tool-calling accuracy.
-
Qwen’s new image model is open-weight, not open-licensed
Alibaba’s Qwen team has published a 7B image generation and editing model with native transparency, but the weights carry a research licence…