OpenAI Cut Luna 80% in One Step. Anthropic Has Cut 20% at a Time.
A.I. / news
OpenAI Cut Luna 80% in One Step. Anthropic Has Cut 20% at a Time.
Since June 26 both labs have repriced their model lines, and the dated record shows two different methods, with DeepSeek still far below either.
OpenAI cut the API price of GPT-5.6 Luna by 80% on July 30, from $1 to $0.20 per million input tokens, while Anthropic's cheapest move on its flagship line has been a 20% step, from Claude Opus 5 to Opus 5.5 on September 22.
The two price histories were laid out on October 25 by Max Woolf on his blog, minimaxir.com, which tracks the changes from June 26 to September 29. The Terminal checked the Anthropic figures against Anthropic's own pages and the OpenAI figures against an OpenAI announcement. OpenAI's launch pages for GPT-6 and GPT-6.1 returned an access error when fetched, so those prices rest on Woolf's table.
Where OpenAI cut and where Anthropic did not
OpenAI's own developer forum confirms the Luna move. A post dated August 21 by an OpenAI account named VeitB lists GPT-5.6 Luna at $0.20 input and $1.20 output per million tokens, and GPT-5.6 Terra at $2 and $12, both described as permanent from July 30.
The same post cut GPT-5.6 Sol by "over 20%" for "the next 3 months", to $4 input and $20 output. That is a promotion with an end date, which a reply in the thread puts at no earlier than November 21, 2026.
Anthropic changed the price of Sonnet 5 without lowering a list figure. Its June 30 launch listed $2 input and $10 output as introductory pricing. A changelog entry on August 10 made that permanent and said the $3 and $15 rate planned for September 1 "no longer applies".
Opus 5.5 followed on September 22 at $4 and $20, against $5 and $25 for Opus 5. Anthropic said Opus 5.5 "requires less compute to serve than Opus 5, and its pricing reflects that", and that it costs about 40% less to run on typical workloads because it uses fewer tokens per task.
- Claude Opus 5.520 USD
- GPT-6.1 Sol10 USD
- Claude Sonnet 510 USD
- GPT-6 Luna0.5 USD
- DeepSeek V4.1 Flash0.6 USD
Source: minimaxir.com, October 25, 2026; Opus 5.5 and Sonnet 5 figures match Anthropic's pages
What the gateway data shows
Vercel's AI Gateway index for October reports that the average price per token across its traffic fell 47% from December to September, while spend rose about 1,180%. It also reports that Anthropic's share of spend dropped from 63.7% to 43.9%.
Open-weight models carried more than seven in ten gateway tokens, and DeepSeek's models alone handled 58% of tokens for 6.6% of spend, according to Vercel. Nearly one in two teams changed their most-used model in September.
That matches Woolf's reading of the Luna cut. He suspects it was aimed at small open-weight models such as DeepSeek Flash, whose cache reads cost $0.003 per million tokens in his table. He also notes that OpenRouter lists 29 providers for DeepSeek V4.1 Flash 1017, up from 13 for V4 Flash.
The Terminal previously covered DeepSeek's V4.1-Flash-1017 release, whose benchmark gains were all self-reported, and Cerebras's walkthrough of Kimi K3, which came with no price.
What is not known
Neither company has said why it chose its method. Woolf raises the possibility that both companies, which he describes as nearing IPOs, are cutting despite the lower revenue, and says he will read Anthropic's prospectus for inference margins. That is his question, not a finding.
OpenAI did not publish a reason for the Luna cut in the forum post. The Sol promotion is scheduled to run until at least November 21, and the Terminal will check the price again then.
Sources
More in A.I.
- 01Qwen3.8-27B Ships Under Apache 2.0 With 262,144-Token Context, While Its 2.4T Sibling Carries a $50 Million Revenue ClauseThe model card lists 25 benchmark rows, all Alibaba's own, and says nothing about training data or the hardware needed.
- 02Microsoft's Decision-1 Is a Post-Trained Qwen3.5-9B at $0.042 Per Million Tokens, With Self-Reported BenchmarksThe model scores fixed answer options instead of writing text, and the launch post does not say whether the weights can be downloaded.
- 03Gemini 4 Argon Reached General Availability October 13 at a $2 Promotional Price, After a Cyber-Partners-First LaunchGoogle's own post gives four benchmark scores, concedes two coding tests to rivals in Reuters' account, and sets a $4 input price from November 30.
- 04Eka's 1,500 Picks an Hour Is a Lab Number, and Its Own Investors Say SoThe Cambridge, Mass. startup raised $150 million for a vision-force-action robot model. Its headline throughput comes from induction and bin-picking tests, and customer pilots are unnamed.