Claude Opus 5.5 vs GPT-6 Sol and Luna: The AI Price War Turns Hot

Two Releases, One Morning: What Just Happened

On September 22, 2026, Anthropic and OpenAI shipped new models within hours of each other — and for the first time in this AI cycle, the headline wasn't raw capability. It was price. Anthropic introduced Claude Opus 5.5, the first model in its new Claude 5.5 family, which performs at the level of its flagship Fable 5.1 on most work while costing roughly 40% less to run than Opus 5. OpenAI answered same-day with GPT-6 Sol and GPT-6 Luna, offshoots of the GPT-6 Astra model launched earlier this month, with API pricing 50% below the promotion it had been running on GPT-5.6.

The synchronized timing is the story. These are the two most-watched AI labs in the West, both fresh off very public pledges to slow down frontier development — and both chose the same morning to compete aggressively on cost. As Fortune put it: "What AI slowdown?" The frontier may be pacing itself, but the fight for everyday business workloads has never been faster.

Claude Opus 5.5: Flagship Performance, 40% Less to Run

Opus 5.5 is the opening release of the Claude 5.5 family, with Sonnet 5.5 and Haiku 5.5 expected "over the coming weeks." Anthropic's pitch is efficiency: the model delivers near-flagship output while requiring less compute to serve, and the company says it is "passing these efficiency savings on to our customers in the form of price cuts and rate limit increases."

• Performance: Early testers reported a 680,000-line code migration completed in under a day — work that would take an engineering team weeks. Asked to cut load times across every page of a web app, Opus 5.5 succeeded 39 of 40 times without altering app behavior, where Opus 5's fixes often changed how the app worked.
• Pricing: $4 per million input tokens and $20 per million output tokens — 20% less than Opus 5. Cache reads drop 60% to $0.20 per million, which matters enormously because caching dominates the bill in agentic and coding workloads.
• Speed: Output generation is more than 30% faster than Opus 5, and five-hour usage limits are going up on Pro, Max, and Team plans.
• Safety: This is Anthropic's first release since CEO Dario Amodei's call to "pace the frontier." The model was tested before release by external evaluators including Frontier Design and METR, scored the best results to date on Anthropic's automated behavioral audit, is more resistant to prompt injection than Opus 5, and ships with the same biology and cybersecurity safeguards as Fable 5.1 — including access for vetted researchers through the Life Sciences Verification Program.

GPT-6 Sol and Luna: Astra's Cheaper Siblings

OpenAI's counterpunch extends the GPT-6 family it opened with Astra on September 3 — the launch President Greg Brockman greeted with "Welcome to the AGI era." Sol and Luna are positioned as versions of Astra tuned for everyday work: the intelligence of the flagship family, tuned for cost and latency rather than maximum capability. OpenAI attributes the price cuts to "improvements in caching and inference" and, like Anthropic, says it is "passing those savings directly onto users and customers."

The playbook is familiar — OpenAI did the same with GPT-5.6, releasing lower-priced variants after the flagship debut. What's different now is the aggression of the cuts and the pressure behind them: cost-conscious enterprises, and a rising tide of competitive open-weight models from China that have reset expectations for what a dollar of intelligence should buy.

Pricing Head to Head

On published numbers, OpenAI's new models undercut Anthropic's on raw API price — but Opus 5.5's $0.20 per million cache reads may be the most consequential figure on the board. In real-world coding agents and automation pipelines, cached tokens can be the majority of total usage, and a 60% cut there reshapes the monthly bill more than headline pricing does. If you're comparing tools, price the workload, not the token list price.

What AI Slowdown? The "Pace the Frontier" Puzzle

Less than two weeks ago, the CEOs of Anthropic, OpenAI, Google DeepMind, Microsoft, and xAI made a rare joint call to slow the development of increasingly capable systems. Amodei published a full pacing essay; Altman publicly backed it. Today's releases don't contradict that — neither Opus 5.5 nor Sol and Luna is a major capability leap past existing frontier models. But they show the pact has a precise boundary: the frontier paces; everything underneath it fights.

"OpenAI and Anthropic are engaged in a price war that is driving down the price of AI and therefore driving down their ability to profit from it," Ramp lead economist Ara Kharazian told Fortune — noting that cheaper models plus outright price cuts on flagships is how technology markets have always worked, "but that is not how AI bulls are pricing it."

The demand side is forcing the issue. Consultants report CFOs hitting AI sticker shock and heading into 2027 planning asking how to optimize costs, with growing adoption of a "cost per completed task" metric rather than cost per seat or per token. When a model gets 40% cheaper and 30% faster at the same time, that math changes overnight — and every app built on top of it inherits the savings.

What It Means for the Tools You Use

• Coding agents are the big winners. A 680k-line migration in a day, 60% cheaper cache reads, and faster output make agent-driven development dramatically more affordable per task.
• Your subscription goes further. Anthropic is raising five-hour usage limits on Pro, Max, and Team plans — expect competitors to match rather than lose power users.
• App prices should fall. Tools built on Claude or GPT APIs just saw their biggest input cost drop. If your favorite AI tool doesn't adjust pricing or limits in the next quarter, ask why.
• Safety is now a shipping feature. External pre-release evaluation (METR, Frontier Design) is becoming a competitive differentiator, not just compliance theater — worth checking for when you choose an AI provider.
• More is coming. Sonnet 5.5 and Haiku 5.5 land in weeks, and the open-weight challengers keep pressing from below. The price war is structural, not a promotion.

Frequently Asked Questions