Sep 20, 2026
OpenAI's newly released GPT-6 Astra model independently beat Valve's Portal puzzle game in under 24 hours at a cost of $571.18 in tokens, with developer cozyblaze publishing the code on GitHub. Nvidia CEO Jensen Huang responded by declaring that AGI has arrived.
Nvidia has put its Groq 3 LPX racks into full production following its $20 billion acquisition of Groq, with the SRAM-based language processing units achieving 3,400 tokens per second and Samsung manufacturing the 4-nanometer chips. The first deployment is set for Nebius Token Factory before year-end.
China's major technology and telecommunications companies are heavily investing in AI infrastructure, with Alibaba reporting a 75% drop in quarterly profit due to increased AI capital expenditure, while state-owned telecoms focus on billable AI tokens as a key growth strategy amid accelerating AI adoption across the country.
Despite Anthropic's strict geoblocking and verification controls, Chinese developers are accessing Claude AI tokens through gray market "transfer stations" at heavily discounted prices, while crypto exchange OKX has banned Claude use for Hong Kong staff after a brief account suspension.
AT&T and other major corporations are increasingly adopting open-weight AI models over proprietary alternatives to reduce costs, with AT&T processing 45 billion AI tokens daily. The shift comes as American tech firms respond to competition from cheaper Chinese open AI models that rival offerings from Anthropic and OpenAI.
Shaw Walters, founder of Eliza Labs, announced that the ELIZAOS token is 'dead' and the Eliza Foundation is winding down after using remaining treasury funds to settle a class-action lawsuit. The token, formerly known as AI16Z, fell 19% to an all-time low following the announcement.
Shaw Walters, founder of Eliza Labs and the ElizaOS AI agent framework, announced the project's token is 'dead' and the foundation is winding down after using remaining treasury funds to settle a Burwick Law class-action lawsuit. The token, formerly known as ai16z, has crashed 97% from its $2.4-2.5 billion peak market capitalization to approximately $305,000.
SpaceXAI released Grok 4.5, its first model since going public, trained in collaboration with Cursor and priced significantly lower than competing models from Anthropic and OpenAI. While trailing GPT-5.5 and Fable 5 in coding benchmarks, the model uses 4.2 times fewer tokens than Opus 4.8 and costs a fraction of competitors.
Microsoft and other tech giants are experiencing a backlash from their push for widespread employee AI adoption, with internal data revealing that AI agents consume significantly more resources—up to 1000x more tokens—than standard AI tools, making them more expensive than human employees. The phenomenon, dubbed 'tokenmaxxing,' has prompted Microsoft to cancel some AI tool licenses.
Anthropic's newly released Claude Opus 4.7 uses a new tokenizer that breaks text into up to 47% more tokens than its predecessor, resulting in significantly higher costs per request despite maintaining the same per-token pricing as Opus 4.6.