Local
OpenAI’s GPT-6.2 Astra Sweeps the Benchmarks — and Power Users Call It ‘Essentially Unusable’
A Roundtable is a way to meet new people and discuss this article in person at a local coffee shop or other venue.
Roundtables are designed for groups of four, with a shared subject and room for everyone to speak. Come for discussion, with no dating expectation.
Source: Cherry Creek News

Cherry Creek News — OpenAI on Tuesday released GPT-6.2 Astra and GPT-6.2 Sol, two models that the company says post new highs across nearly every benchmark it published — while costing 20 percent less per token than the flagship they replace. Within hours, some of the most influential users of these models publicly called the release a failure. Investor and former HyperWrite CEO Matt Shumer wrote that the models “were essentially unusable.” Developer and YouTuber Theo Browne was blunter: “the worst model OpenAI has ever released.” It is an unusually sharp split for a launch that, on paper, looks like a clean sweep. OpenAI’s own release notes claim GPT-6.2 Astra sets a new high on Agents’ Last Exam, scoring 5.2 points above GPT-6 Astra at less than a third of the cost, and 5.5 points ahead of Anthropic’s Claude Fable 5.2 at roughly 5 percent of the cost. Independent evaluators broadly confirm the gains — and just as broadly confirm the complaints.
This excerpt comes from the named source. Read the original for the full reporting.
Read the original at Cherry Creek News


