Anthropic and OpenAI Both Cut AI Prices the Same Day, and the Timing Isn’t a Coincidence
Claude Opus 5.5 and GPT-6 Sol and Luna landed within hours of each other on Tuesday, each promising frontier performance at roughly half the old cost, a squeeze that hits agent-heavy developers hardest.
Anthropic and OpenAI each released new models on Tuesday, and this time the marketing pitch was not a leap in raw intelligence, but a sharp cut in what it costs to run intelligence the industry already has.
Anthropic’s Claude Opus 5.5 performs at the level of its Claude Fable 5.1 on most work while costing 40% less than its predecessor, Opus 5, the company said.
Hours later, OpenAI introduced GPT-6 Sol and Luna, cheaper siblings to its flagship GPT-6 Astra, cutting API prices 50% versus GPT-5.6’s promotional rates.
Neither company called the other out directly, but the overlap in timing, target customer, and message was hard to miss.
Two Different Paths to the Same Discount
As expected, Anthropic dropped Opus 5.5 and priced it at $4 per million input tokens and $20 per million output tokens, a 20% reduction from Opus 5’s $5 and $25 rates.
The bigger saving sits in cache reads, which make up most agentic coding costs and drop 60% to $0.20 per million tokens.
OpenAI took the opposite route: rather than shave one flagship model’s price, it expanded the GPT-6 lineup itself, cutting Sol’s input price from $4 to $2 and Luna’s from $0.20 to $0.10 per million tokens, each a straight 50% reduction.
On OpenAI’s own AutomationBench figures, GPT-6 Sol at its highest effort setting outscored Claude Opus 5 at max effort while costing roughly 9% as much per task, a gap Anthropic’s new pricing seems to close.
Why the Discount Lands on Agent Builders First
Both companies pointed to the same customer: developers running multi-step agents that burn millions of tokens, rather than standard chatbots handling turn-based replies.
Anthropic said an early tester used Opus 5.5 to audit and fix a 200,000-line codebase in under three hours, work that took Opus 5 more than 20 hours and 2.5 times the tokens.
OpenAI, similarly, said its median researcher now burns more than $600 a day in Codex token usage at API prices, with heavy users past $7,000, a number that makes a 50% price cut consequential rather than cosmetic.
GitHub told OpenAI that improved prompt caching has already cut the share of tokens needing fresh processing by more than half across billions of requests.
For any team billing by the token, Tuesday’s releases were less about smarter models and more about which company can now serve roughly the same intelligence for less compute.
The Discount War Has a Safety Subplot Underneath It
What makes this price war different from past clashes is what’s sitting next to it.
Opus 5.5 is Anthropic’s first release since CEO Dario Amodei called to pace frontier capabilities against safety, and it ships with Fable-level cyber and bio safeguards while rerouting select cyber tasks to Opus 4.8.
This creates an obvious tension: Anthropic is racing OpenAI on price while advocating for industry slowdowns, even as Opus 5.5’s system card notes the model often seems to sense when it’s being evaluated.
And this is a limit Anthropic says makes real-world behavior harder to predict.
OpenAI’s Sol and Luna, meanwhile, ship as extensions of GPT-6 Astra’s alignment work but without Astra’s own top-tier restrictions.
The practical result is that the models most people will actually use, the cheap, high-volume workhorses, are getting less individual scrutiny than the flagship models built to headline a safety pitch.
Cheaper AI is good news for anyone’s budget; it says nothing about whether the safety promises attached to the expensive version travel down to the model everyone’s agents are actually running on.
Source(s): We’re introducing Claude Opus 5.5
