Anthropic and OpenAI Launched Rival Models 90 Minutes Apart, and Then Cut Prices in Half

Anthropic released Claude Opus 5.5 on 22nd September, and OpenAI followed with GPT-6 Sol and GPT-6 Luna about an hour later, Simon Willison writes. Both companies cut prices at the same time, and the releases landed just over a week after Anthropic co-founder Dario Amodei published an essay arguing that AI labs need to slow down and "pace the frontier" rather than race each other. This piece covers what each company shipped, what the numbers actually show, and what a price war between the two biggest AI labs means for anyone building with their tools, in Nigeria or anywhere else.

What Anthropic Released

Claude Opus 5.5 is Anthropic's new flagship model, priced at $4 per million input tokens and $20 per million output tokens, a 20 percent cut from Opus 5's $5 and $25, Simon Willison reports, with cached input token costs down 60 percent to $0.20 per million. Anthropic engineer Thariq Shihipar described the model as one that "communicates clearly," addressing complaints about Opus 5's tone, while carrying "the intelligence of Fable 5.1" at a lower cost per token than its predecessor.

Willison also flags one rough edge: pushed to its maximum thinking setting while generating a test SVG image of a pelican, Opus 5.5 ran past its own 128,000-token output limit, suggesting the highest thinking setting is not yet practical for heavier tasks.

Source: Anthropic

What OpenAI Released, and How the Timing Lined Up

GPT-6 Sol and GPT-6 Luna, Decrypt reports, are faster, cheaper alternatives to OpenAI's existing GPT-6 Astra, rolled out across ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Education subscribers within roughly an hour of Anthropic's announcement. OpenAI says the two models "build on the advances behind GPT-6 Astra, bringing much of its strengths into faster and more affordable models to support work at scale."

Sol costs $2 per million input tokens and $10 per million output tokens, half of GPT-5.6's $4 and $20, Decrypt notes. Luna is cheaper still, at $0.10 per million input tokens and $0.50 per million output tokens, down from $0.20 and $1.20, making it one of the least expensive models OpenAI has ever released. OpenAI's own benchmark claim has Sol scoring 33.2 percent on AutomationBench at $0.27 per task, against Claude Opus 5's 26.9 percent at more than eleven times the cost per task, a comparison that uses the older Opus 5 rather than the same-day Opus 5.5.

Source: OpenAI

Why the Timing Was Not a Coincidence

Neither company has said the releases were coordinated, but the pattern is now familiar: whenever either lab ships a major model, the other tends to answer within days, and this time the gap closed to under two hours, Simon Willison notes. The releases also followed a wider one, with xAI's Grok 4.7 and MiMo v2.6 Flash and Pro arriving a day earlier on 21st September. Willison points out that Grok 4.7 is now priced identically to GPT-6 Sol on input tokens, and that GPT-5.6 Terra, OpenAI's own recent model, became effectively redundant the moment Sol matched its price at a newer generation.

The releases came nine days after Amodei published an essay, "We Must Pace the Frontier," arguing that the industry's compressed release cycle carries real risk and that AI labs should slow down rather than race to ship. Commentary on the essay, including from TechPolicy.Press, questioned whether Anthropic itself was positioned to be the one setting that pace. The same-day launch race that followed drew renewed attention to that tension between what labs say about safety and how quickly they actually ship.

What the Price Cuts Mean for Developers

The immediate effect is that access to near-frontier AI got substantially cheaper across the board. AI Magicx describes 2026 as a year in which AI costs have dropped as much as tenfold from where they stood at the start of it, driven by exactly this kind of back-to-back undercutting between labs. For a Nigerian developer, startup or small business paying for AI access in dollars, a 50 to 60 percent price cut on a flagship model changes the calculation on what is affordable to build, particularly for anything that calls a model repeatedly rather than once.

That matters beyond hobby projects. PINCHED has covered how Nigeria stood out as the one country in a 37-nation Pew Research survey where people said AI was more likely to help their careers than hurt them. Cheaper access to frontier-level models is one of the more direct ways that optimism could translate into actual use, since cost has been one of the main barriers keeping smaller developers and businesses outside Nigeria's biggest tech hubs from building with the newest models at any real scale.

What to Watch Next

Whether this round of cuts holds, or gets undercut again within weeks, is the open question. The pattern all through 2026 has been that no price cut stays the lowest for long, and Grok 4.7 landing at the same input price as GPT-6 Sol on the very next day suggests the next move is already in motion. There was no obvious interview subject on this story who could speak on the record ahead of publication, since it concerns internal pricing and release decisions at two major AI labs; PINCHED will follow up if either company comments further.

This article is a news report, and PINCHED has not been briefed directly by Anthropic or OpenAI.

Frequently Asked Questions

What Is Claude Opus 5.5?

Claude Opus 5.5 is Anthropic's newest flagship AI model, released on 22nd September 2026, priced at $4 per million input tokens and $20 per million output tokens, a 20 percent reduction from Opus 5.

What Are GPT-6 Sol and GPT-6 Luna?

GPT-6 Sol and GPT-6 Luna are faster, cheaper OpenAI models released the same day as Claude Opus 5.5, built to bring much of the capability of GPT-6 Astra into more affordable models. Sol costs $2 per million input tokens and $10 per million output tokens; Luna costs $0.10 and $0.50.

Why Did Anthropic and OpenAI Release Models on the Same Day?

Neither company has confirmed coordination, but the two labs have a pattern of answering each other's major releases within days, and on 22nd September the gap closed to about an hour.

What Is Dario Amodei's "Pace the Frontier" Essay About?

Published on 13th September 2026, the essay argues that AI labs are moving too fast and that the industry should slow its release cycle for safety reasons, a position that drew scrutiny given Anthropic's own release pace days later.

What Does This Price War Mean for Developers Outside the US?

Sharp price cuts on flagship models lower the cost of building with frontier AI for developers and businesses anywhere, including in Nigeria, where cost has been one of the practical barriers to wider adoption of the newest models.