Anthropic, Google and OpenAI All Line Up Next-Generation Models, Report Says
A report from The Information published on October 19 says Anthropic, Google and OpenAI are each preparing new frontier models, with Anthropic's unreleased system singled out for its ability to infer what users actually intend. The three-way race sets up a crowded stretch of launches, even as investors grow choosier about the AI trade.
The next round of frontier AI releases is already taking shape. A report published by The Information on October 19 says Anthropic, Google and OpenAI are each advancing work on next-generation systems, setting up a crowded stretch of launches across the three labs that have dominated the top of the market this year.
A three-way race on the runway
The report, which draws on people familiar with the companies' plans and on developers who have tested early versions, describes three separate efforts rather than a single coordinated release window. Each company sits at a different stage of readiness, and none has publicly confirmed a launch date.
Anthropic's model: intuition over instructions
The most detailed claims concern Anthropic. According to The Information, the company is developing a model with stronger reasoning and what the report describes as improved intuition — the ability to work out what a user is actually trying to achieve from thin or incomplete instructions.
One person who tested the system told the publication that it can look past an individual software bug and identify the broader problem behind it, then propose a more comprehensive fix. That is a meaningful shift from the pattern-matching behaviour that has limited coding assistants in practice: patching the line that broke rather than the design that allowed it to break.
The report did not disclose a name or a firm release date for the Anthropic system. The omission is notable given that the company has been weighing a model release ahead of its long-anticipated public listing; Reuters reported in September that Anthropic was considering exactly such a move to counter momentum OpenAI built with its GPT-6 Astra launch.
Google lines up Gemini 4 Carbon
Google's effort looks like a refinement rather than a new generation. The Information reports that the company is working on Gemini 4 Carbon, an upgrade to Gemini 4 Argon, which Google announced on September 30 and began rolling out in October.
Argon arrived with a 1-million-token context window and a pitch built around long, multi-step professional work — coding, financial research, legal drafting and autonomous vulnerability discovery. Developers who spoke to The Information said Google's models remain particularly strong on visual tasks, including video analysis and website testing, a niche where its multimodal training gives it an edge.
OpenAI's Terra
OpenAI, meanwhile, appears to be preparing a model that has circulated in reporting under the name Terra, understood as a follow-up in the GPT-6 line that produced Astra. Details are thinnest here, which is consistent with OpenAI's recent habit of keeping release information tightly controlled until launch.
Why the timing matters
The clustering is not coincidental. Model releases have become competitive signalling as much as product events: a launch resets benchmarks, shifts developer attention and, for a company heading toward a public listing, hands investors a fresh narrative.
It also lands against a more sceptical backdrop. Investors have grown choosier about AI-adjacent offerings — data-centre listings in particular have stumbled — and each lab now faces harder questions about whether the next model represents a genuine capability jump or an incremental tune.
What we still don't know
None of the three companies has confirmed the report. Names, dates, pricing and benchmark results are all unverified, and reporting of this kind has often preceded launches by weeks or months — sometimes with different branding entirely. Treat the specifics as a signal of direction rather than a schedule.