OpenAI and Anthropic Ship Rival AI Models 90 Minutes Apart, Both Cutting Prices
Two Launches, Ninety Minutes Apart
On September 22, 2026, OpenAI and Anthropic each announced major updates to their flagship AI model lines within roughly 90 minutes of one another. Both companies also cut prices on their respective offerings the same day. The near-simultaneous timing quickly became a talking point in tech press coverage, though it's worth noting that the competitive-timing narrative largely originates from how journalists framed the announcements rather than any confirmed strategic coordination or intent by either company.
OpenAI's GPT-6 Sol and Luna: What Was Announced
OpenAI introduced two new models, described in press coverage as GPT-6 Sol and Luna, positioned as more efficient successors within its GPT-6 lineup. According to OpenAI, GPT-6 Sol is available at roughly half the cost of its 5.6-series predecessor. OpenAI also says the new model makes about half as many factual errors, based on its internal evaluation process. The company states the models are rolling out across ChatGPT Work, Codex, and its API, with plans to expand to Free and Go tiers.
It's worth flagging that OpenAI's own official announcement page was not accessible for direct review, meaning much of the naming and detail here relies on secondary press coverage rather than primary confirmation. As with most company product announcements, the performance and accuracy figures cited are self-reported and have not been independently benchmarked.
Anthropic's Opus 5.5: What Was Announced
Anthropic announced Claude Opus 5.5, cutting output token pricing to $20 per million tokens, down from $25 for its predecessor. The company says Opus 5.5 matches or exceeds the performance of a larger internal model it refers to as "Fable," though this comparison is difficult to evaluate independently since "Fable" is not a publicly benchmarked or widely defined reference point outside Anthropic's own materials.
Notably, Anthropic published a system card alongside the release and pointed to a pre-deployment safety evaluation conducted by the outside research organization METR. The company also cited another external group, Frontier Design, in connection with biological and cybersecurity risk review. Anthropic further stated that Sonnet 5.5 and Haiku 5.5 versions are expected in the coming weeks.
While the third-party safety review is a meaningful point of external validation, it's distinct from performance verification. The comparative capability claims — including the "Fable-level performance" framing — remain based on Anthropic's own reporting.
Competing Narratives: Speed vs. Caution
The two companies' public messaging reflects noticeably different framing. OpenAI's announcement leans into cost savings and accuracy improvements, using language that echoes typical product marketing. Anthropic, meanwhile, frames its release around a stated policy of pacing capability advances alongside safety and alignment work, with CEO Dario Amodei's comments lending a more deliberative, risk-conscious tone to the announcement.
These are contrasting public postures rather than independently verified strategies. Many observers note that both framings serve each company's broader narrative — OpenAI emphasizing speed and value, Anthropic emphasizing prudence — but neither approach has been substantiated by outside analysis of the underlying business or safety tradeoffs.
What's Actually Verified — and What Isn't
Stepping back, it's worth being clear about what can and cannot currently be confirmed. Both companies' claims about accuracy, error rates, and comparative performance originate from internal, self-reported evaluations. No independent third-party benchmarking appears in the available sourcing for either model at this time.
Anthropic's citation of METR represents a genuine external safety check, but this is separate from verifying performance or capability claims. On OpenAI's side, the specific naming and details of "GPT-6 Sol and Luna" remain unconfirmed by OpenAI's own primary announcement page, which was not accessible for direct review; the available detail relies on secondary tech press summaries. A recurring consumer concern in coverage like this is that headline comparisons — such as matching "Fable" or halving error rates relative to a prior model — are company statements that merit some skepticism pending independent confirmation.
Why This Matters
Simultaneous price cuts from two of the leading AI developers suggest the competitive pressure in the frontier model market continues to intensify. The episode is also a useful illustration of how AI product announcements are frequently relayed by press largely as-is, with limited independent verification of the underlying technical claims. As rivalry narratives around timing and positioning draw attention, it's worth keeping sight of the distinction between confirmed facts and company-supplied claims awaiting substantiation.