OpenAI and Anthropic Just Cut Prices by Half. The AI Price War Is Here.

September 22 was not a normal day in the AI industry. Within hours of each other, OpenAI and Anthropic both released new flagship models with a shared headline: prices are dropping by 40 to 50 percent, and the models are better than what came before.

OpenAI shipped GPT-6 Sol and GPT-6 Luna. Sol costs $2 per million input tokens and $10 per million output tokens. Luna, the lighter variant, runs at $0.10 and $0.50 respectively. Both are roughly half the price of their GPT-5.6 counterparts. Anthropic answered with Claude Opus 5.5, priced at $4 and $20 per million tokens, which is 40 percent cheaper than Opus 5 while reportedly matching the performance of Claude Fable 5.1.

What makes this unusual is not just the scale of the cuts. It is the timing. Two companies that usually stagger their releases to avoid direct comparison dropped competitive pricing on the same day. Industry watchers like Simon Willison immediately called it what it looks like: a price war.

Why This Is Happening Now

The timing follows weeks of tension over AI safety and development speed. Anthropic CEO Dario Amodei published an essay on September 12 urging the industry to slow down, and several AI lab leaders publicly agreed. A lawsuit filed days later alleges that OpenAI, Anthropic, SpaceXAI, and Google illegally coordinated to slow development. Against that backdrop, launching faster, cheaper models is a strong signal that competitive pressure is winning over the slowdown camp.

There is also a technical driver. Both companies credit improvements in inference infrastructure and prompt caching. OpenAI specifically notes that better caching lets the models reuse more context and respond faster, which lowers the cost to serve. Anthropic says Opus 5.5 requires less compute to run than its predecessor. In other words, these are not loss-leader discounts. The companies believe they can maintain margins at lower prices.

What the Benchmarks Say

The price drops come with genuine capability jumps. OpenAI claims GPT-6 Sol makes about half as many mistakes as GPT-5.6 Sol on internal factuality tests and matches Claude Fable 5.1 on coding benchmarks at a fraction of the cost. On FrontierCode, which grades whether coding agents produce changes ready to merge into real codebases, Sol is competitive with Fable 5.1. On AutomationBench, a test of business workflows across 47 tools, Sol at xhigh effort outperforms Claude Opus 5 at max effort at roughly 9 percent of the cost per task.

Anthropic’s Opus 5.5 leads on Terminal-Bench 4.0 at 66.4 percent, ahead of GPT-6 Astra’s reported 57.9 percent, and scores 54.4 percent on FrontierCode v1.1. Anthropic also emphasizes safety metrics: Opus 5.5 allegedly attempts to circumvent boundaries 85 percent less often than prior models and resists prompt injection better than Opus 5. It is the highest-scoring model to date on Anthropic’s automated behavioral audit.

What This Means for Developers

For people actually building with these APIs, the practical impact is straightforward. Costs for complex agentic coding, long-context workflows, and automated business processes just fell by half or more. OpenAI noted that its internal researchers now burn through over $600 in API tokens per day at median usage, with power users hitting $7,000. Cutting those bills in half changes what teams can afford to automate.

The ecosystem responded immediately. GitHub Copilot added Opus 5.5 across its stack the same day. Devin reported a new number-one result on FrontierCode at less than a tenth of Fable 5’s cost. AWS Bedrock, Google Cloud, Azure, and Databricks all added one or both model families within hours.

The Broader Context

This wave of releases, which also includes Grok 4.7 and MiMo v2.6, points to a market that is both accelerating and commoditizing. The frontier is still moving forward, but the economics of using frontier-level models are improving faster than many expected. If the trend holds, the competitive moat for AI-native applications will shift from access to the best model toward execution, integration, and user experience.

That is good news for builders and bad news for anyone hoping the industry would collectively pump the brakes. The price war is here, and the next few months will show whether these cuts are sustainable or just the opening salvo in a race to the bottom.