AI
Anthropic and OpenAI Cut Prices on the Same Tuesday, and the Cheapest Thing Became Memory
On Tuesday, September 22, the two biggest AI labs cut prices within hours of each other, and the coincidence stopped at the calendar. Anthropic introduced Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. OpenAI answered with GPT6 Sol at $2 and $10 and GPT6 Luna at $0.10 and $0.50. Sol lands at exactly half of Opus 5.5, the same rate Anthropic already charges for its own cheaper Sonnet 5 tier. OpenAI priced its new model against a rival product, on the day that product's own price changed.
The headline numbers only tell part of the story. Anthropic's deepest cut went to cache reads, the price of reusing context an agent has already processed, which fell 60 percent from $0.50 to $0.20 per million tokens. Anthropic says a typical workload now costs 40 percent less to run than on Opus 5, because the model is cheaper per token and uses fewer tokens per task. It also generates output more than 30 percent faster than Opus 5 and, on Anthropic's own testing, performs at the level of Claude Fable 5.1 on most work. For coding agents and long conversations that reread the same context every run, the cache number moves the bill more than the headline rate.
OpenAI's pair sits below GPT6 Astra, the flagship the company introduced earlier this month. The company calls the move a 50 percent cut compared with GPT5.6 promotional pricing and credits improvements in caching and inference. Luna's output price falls even further, from $1.20 to $0.50. An OpenAI spokesperson confirmed to VentureBeat that both rates are permanent list prices, set for the long term. Both models are available in ChatGPT Work and in Codex for Plus, Pro, Business, Enterprise and Edu users, with Luna also in the desktop app for Free and Go plans. Developers reach both through the API.
The benchmark tables deserve a careful read, because both were written on the day both prices changed. OpenAI's published comparison measures Sol against Claude Opus 5, the model Anthropic just replaced. Anthropic's measures Opus 5.5 against GPT6 Astra, OpenAI's separate top tier. Each company picked the tests that flatter its own model, and neither compares its new model against the other's new model. Anthropic adds a caution worth keeping. At these capability levels, benchmark margins have become a less reliable guide to real world differences. The honest advice for any team with a fixed monthly budget is to rerun its own top workload at the new prices before renewing any enterprise contract signed even a few weeks ago.
xAI read the same market and moved the other way, doubling its rate for long prompts above 200,000 tokens. And the week's deeper story sits behind the token prices. The cost of running AI products is shifting from raw model calls to the systems around them, agent memory, observability, and the platforms that meter every agent action. Cheaper tokens make the agent economy bigger. The companies that own the agent's tools, its memory, and its audit trail are where the bill is heading.
For anyone building on these models, Tuesday was a gift with homework attached. Your per token costs just fell by a third to a half, and your vendor comparisons are already stale. Run the numbers on your own workloads, keep an eye on the cache line where the real savings live, and budget for the layer above the model, because that is where the next pricing battle is already taking shape.
Quick answers
What is this story about?
On Tuesday, September 22, the two biggest AI labs cut prices within hours of each other, and the coincidence stopped at the calendar. Anthropic introduced Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens, down from $5 and $25 for Opus 5. OpenAI answered with GPT6 Sol at $2 and $10 and GPT6 Luna at $0.10 and $0.50. Sol lands at exactly half of Opus 5.5, the same rate Anthropic already charges for its own cheaper Sonnet 5 tier. OpenAI priced its new model against a rival product, on the day that product's own price changed.
Why does this story matter?
For anyone building on these models, Tuesday was a gift with homework attached. Your per token costs just fell by a third to a half, and your vendor comparisons are already stale. Run the numbers on your own workloads, keep an eye on the cache line where the real savings live, and budget for the layer above the model, because that is where the next pricing battle is already taking shape.
Sources
- MRKR: Anthropic and OpenAI cut AI model prices on the same day
- Servola Journal: OpenAI Priced Its New Model at Half of Anthropic's
- AI Explainer India: AI Price War, Claude Opus 5.5 vs GPT-6 Sol and Luna
New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.