Morning Edition · №
AI Models SAN FRANCISCO

OpenAI Ships GPT-6.2 Astra and Sol as Microsoft Fires Back With Rival Coding Models

The dueling releases underscore how quickly the frontier-model race has shifted toward cheaper, task-specific tools aimed at professional work rather than headline-grabbing general intelligence.

SHARE X f in ⧉

OpenAI this week released GPT-6.2 Astra and Sol, two new models the company says are tuned for professional work while cutting the computing cost of complex tasks, as Microsoft moved to counter with its own coding-focused releases, MAI Code 2 and a lighter MAI Code 2 Flash.

The near-simultaneous releases mark a shift in emphasis among the largest AI labs, which have spent much of the past two years one-upping each other on general reasoning benchmarks and are now increasingly competing on cost and reliability for specific professional workflows — legal research, software engineering and enterprise data analysis among them.

Specialization over size

OpenAI has already begun applying GPT-6.2 Astra to its own legal-research tooling, the company said, positioning the model as a workhorse for complex, document-heavy tasks rather than a flashy consumer chatbot upgrade. Microsoft's MAI Code 2 is aimed at complex software development projects, while the Flash variant targets routine bug fixes and smaller coding tasks at a lower price point — a split that mirrors a broader industry trend toward tiered model families rather than a single flagship release.

The timing puts both companies back in direct competition just as their own relationship has been shifting: Microsoft and OpenAI recently renegotiated the terms of their partnership, ending Microsoft's exclusive right to resell OpenAI's models and clearing the way for Microsoft to lean more heavily on AI systems it builds in-house.

A crowded, cost-conscious market

The dueling launches land in the middle of a bruising month for AI-linked stocks, with investors increasingly scrutinizing whether the enormous capital spending behind these models is translating into matching revenue. Cheaper, more specialized models are, in part, a response to that pressure: enterprise customers have pushed back on costs, and labs are racing to show they can deliver professional-grade results without the runaway compute bills that come with their largest general-purpose systems.

Neither company disclosed pricing details for enterprise tiers of the new models, though both said broader availability would roll out over the coming weeks.

SHARE THIS ARTICLE X Facebook LinkedIn Copy link
Claire Fontaine · Technology & Regulation Correspondent

Reports on technology and its regulation for UBStandard, with a focus on Brussels, AI policy and Europe's digital economy.

[email protected]
Related coverage Front page →