Industry · Software · Tech Policy · Events
New products, new platforms and the first impressions that matter, reviewed by The Silicon Ledger.
OpenAI distributed its new GPT-6 Astra model to paying ChatGPT subscribers on Friday, one day after its announcement, following CEO Sam Altman's public apology for what he termed a "messy rollout."
OpenAI unveiled GPT-6 Astra on Thursday with published documentation and pricing, but a chaotic rollout prompted CEO Sam Altman to apologize as many developers remain unable to access the new model.
OpenAI claims GPT-6 Astra achieved 98.6% on the ARC-AGI-3 benchmark, a dramatic leap from the 7.8% posted by its predecessor GPT-5.6 Sol. However, undisclosed evaluation parameters and unclear methodology raise questions about what the score actually demonstrates.
OpenAI unveiled GPT-6 Astra on Thursday, calling it the world's most intelligent model and suggesting it marks the beginning of the AGI era. The flagship model delivers strong results on specialized benchmarks but commands higher API costs and does not clearly dominate in coding tasks.
Meta released Muse Spark 1.3 this week, claiming its strongest performance gains yet in coding and agent tasks, while also formally launching its Muse Code agent with new subscription tiers. The new model briefly dethroned Google's freshly released Gemini 3.8 Flash from the cost-efficiency frontier.
Google has released its third Flash model iteration in six weeks, claiming performance parity with Anthropic's Opus 5 while holding the line on introductory pricing.
The Spanish AI company unveiled its first large-scale model, a compressed 438-billion-parameter system designed for coding and enterprise agents, but questions linger about whether benchmark speeds translate to practical deployment.
Google has unveiled Gemini 3.8 Flash alongside a restricted-access Cyber variant, with the standard model approaching Anthropic's Opus 5 on coding benchmarks and the Cyber version limited to 650 trusted security partners through a new Fairwind Program.
