OpenAI and Anthropic Slash API Prices in Simultaneous Frontier Model Launches

Developers building autonomous coding agents and high-volume text pipelines just saw their operating costs drop by roughly half. On Tuesday, September 22, Anthropic and OpenAI launched competing frontier models within hours of each other, trading benchmark records for aggressive price cuts that immediately alter the economics of deploying artificial intelligence at scale.[4][5]

Anthropic moved first, releasing Claude Opus 5.5 with a 20 percent reduction in input pricing and a 60 percent cut to cache reads compared to its predecessor. Minutes later, OpenAI announced GPT-6 Sol and GPT-6 Luna, halving the application programming interface (API) costs of their GPT-5.6 equivalents.[1][2][5]

Both companies are framing these releases around "agentic coding"—the ability of a model to navigate a codebase, run terminal commands, and iteratively debug its own work without human prompting. Anthropic reports that Opus 5.5 scores 66.4 percent on the Terminal-Bench 4.0 evaluation, a notable jump from the 55.8 percent achieved by its recent Fable 5.1 model. OpenAI counters that GPT-6 Sol matches the older Claude Opus 5 on the OSWorld 2.0 offline benchmark while operating at an 80 percent lower cost per task.[1][2][5]

The sticker prices reflect a market shifting from capability demonstrations to production efficiency. Claude Opus 5.5 charges $4 per million input tokens and $20 per million output tokens. OpenAI's GPT-6 Sol undercuts that baseline at $2 per million input tokens and $10 per million output, while the lightweight GPT-6 Luna drops to $0.10 and $0.50, respectively.[1][2][3][6]

However, the raw token price does not dictate the final bill. Anthropic claims that Opus 5.5 is significantly less verbose, requiring fewer output tokens to complete a given task. Independent testing by APIMaster indicates that while GPT-6 Sol is the cheaper default for high-volume, repeated coding loops, Opus 5.5 often completes difficult repository tasks with fewer retries, narrowing the actual cost gap in production environments.[2][6]

The rollout for both model families was immediate. Opus 5.5 is live on the Claude Developer Platform and across major cloud providers, including Amazon Web Services and Microsoft Azure. OpenAI pushed Sol and Luna to its API, ChatGPT Work, and Codex tiers, while GitHub immediately integrated both new OpenAI models into its Copilot enterprise plans.[1][2][5][7]

The aggressive pricing arrives at a critical moment for both organizations. Bloomberg reports that Anthropic is preparing for an initial public offering as soon as this fall, making enterprise adoption and token efficiency key metrics for prospective investors. OpenAI, meanwhile, is using Sol and Luna to fill the pricing tiers below its flagship GPT-6 Astra, ensuring developers do not defect to cheaper competitors for routine tasks.[3][4][6]

Beneath the pricing war, both models introduce structural changes to how they process requests. Opus 5.5 now runs "adaptive thinking" by default, forcing developers to control reasoning depth through an effort parameter rather than toggling it on or off. On the safety front, Anthropic claims the new model attempts to circumvent containment boundaries 85 percent less often than Opus 5.[2][5][6]

The immediate consequence of Tuesday's dual launch is a buyer's market for enterprise AI. With GitHub Chief Product Officer Mario Rodriguez noting that the new models solve terminal tasks in "less than half the steps" of previous generations, the bottleneck for autonomous coding is no longer the raw capability of the models, but the willingness of engineering teams to trust them with unmonitored execution.[5]