Anthropic’s Claude Opus 5.5 lists at $20 per million output tokens, against $75 for Claude 3 Opus at its March 2024 launch: a 73% cut, by our arithmetic. Over the same span, the Opus line’s score on Epoch AI’s Epoch Capabilities Index rose 42.7 points, from 125.63 to 168.36, by the research group’s published data (Epoch AI).
OpenAI’s GPT-6 Astra lists at $50 per million output tokens, five times GPT-4o’s $10, while GPT-6.1 Sol scores within 0.13 index points of Astra at $10, the same output price as GPT-4o, by our arithmetic from the two companies’ pricing pages. Anthropic also lists Claude Fable 5.1 at $50; Epoch scores it 166.9, below Opus 5.5.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing, where customers pay per unit used. It teaches the general idea only and does not discuss any company, model or figure named here.
The key insight: By our arithmetic, the Opus line’s output list price fell 73%, from $75 to $20, while its Epoch Capabilities Index score rose 42.7 points; GPT-6.1 Sol scored 40.7 points above GPT-4o at the same $10 output price.
On Anthropic’s page, Claude Opus 4 still lists at $15 and $75. Claude Opus 4.5 and Claude Opus 5 list at $5 and $25. Claude Opus 5.5 lists at $4 and $20. Claude Fable 5.1, the highest-priced model in Anthropic’s main lineup, lists at $10 and $50, the same as GPT-6 Astra.
On OpenAI’s pages, GPT-4o lists at $2.50 input and $10 output, o3 at $2 and $8, GPT-5 at $1.25 and $10, and GPT-5.5 at $5 and $30. The current pricing page lists GPT-6 Astra at $10 and $50 and GPT-6.1 Sol at $2 and $10.
The scores come from Epoch AI’s Epoch Capabilities Index, which combines results from many benchmarks into one number per model; the file we used was downloaded on November 8. Claude 3 Opus scores 125.63, Claude Opus 4 142.83, Claude Opus 4.5 151.82, Claude Opus 5 164.45 and Claude Opus 5.5 168.36. Claude Fable 5.1 scores 166.9.
On the same index, GPT-4o (the May 2024 version) scores 128.23, o3 146.99, GPT-5 150.0, GPT-5.5 160.04, GPT-6.1 Sol 168.94 and GPT-6 Astra 169.07.
Epoch Capabilities Index score against list output price per million tokens, log scale. Anthropic’s Opus line (ochre) fell from $75 to $20 as its score rose from 125.63 to 168.36; OpenAI’s GPT-6.1 Sol (hollow teal) scores 168.94 at $10, GPT-6 Astra 169.07 at $50. List prices read 8 Nov 2026; Claude 3 Opus at its March 2024 launch price.
Two Pricing Paths
By our arithmetic, Anthropic’s Opus line gained 42.73 index points between Claude 3 Opus and Claude Opus 5.5 while both its input and output list prices fell about 73%, from $15 to $4 and from $75 to $20.
OpenAI’s path split at the top. By our arithmetic, GPT-6.1 Sol gained 40.71 points over GPT-4o at the same $10 output price, with input 20% cheaper at $2. GPT-6 Astra gained 40.84 points at five times GPT-4o’s output price.
At the frontier the three newest models sit close together. Epoch’s 90% credible intervals overlap: GPT-6 Astra 166.16 to 172.28, GPT-6.1 Sol 165.97 to 172.19 and Claude Opus 5.5 165.54 to 171.51. By our arithmetic, Opus 5.5’s $20 output price is 40% of Astra’s $50, and Sol’s $10 is a fifth.
As we reported, Epoch put a 72% chance on GPT-6 Astra scoring above Claude Opus 5.5. The pricing pages show the main lineups of both companies with the same top and bottom list prices, as we reported.
Our Number
By our arithmetic, the Opus line’s output list price fell 73% while its Epoch index score rose 42.7 points; OpenAI’s GPT-6.1 Sol rose 40.7 points at an unchanged $10 output price.
The Structural Read
As we read the two price paths, Anthropic’s Opus output list price has fallen across the line, from $75 for Claude 3 Opus and Opus 4 to $25 for Opus 4.5 and Opus 5 and $20 for Opus 5.5, while OpenAI’s newest scores sit at two price points, $10 for GPT-6.1 Sol and $50 for GPT-6 Astra.
As we read Epoch’s data, the three highest-scoring models, GPT-6 Astra, GPT-6.1 Sol and Claude Opus 5.5, cannot be told apart with confidence, because their 90% credible intervals overlap.
As we read Anthropic’s page, its own $50 model, Claude Fable 5.1, scores 166.9 on Epoch’s index, below the $20 Opus 5.5, so in Anthropic’s lineup the highest list price does not buy the highest Epoch score.
Anthropic, launching Claude 3 Opus at $15 and $75, March 2024
“Claude 3 Opus is our most intelligent model, with best-in-market performance on highly complex tasks.”
Three Implications
FOR BUYERS By our arithmetic, Opus 5.5’s $20 output list price is 40% of GPT-6 Astra’s $50, and GPT-6.1 Sol’s $10 is a fifth, for scores within Epoch’s overlapping intervals.
FOR OPENAI GPT-6 Astra lists at five times GPT-4o’s $10 output price, by our arithmetic, for a 40.84-point higher Epoch score.
FOR ANTHROPIC Claude Opus 4 still lists at $15 and $75, the same as Claude 3 Opus at launch, while Opus 4.5 through Opus 5.5 list at $25 or less for output.
The analysis describes inference this way: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”
As we read it, list prices per million tokens are the per-query price side of that engine. The analysis does not discuss these models or Epoch’s scores.
What Is Not Established
List prices are not what large customers pay; discounts, batch rates and caching are excluded here. The index has no true zero, so the comparison is of changes in score against changes in price, not cost per point. GPT-4o’s price is today’s list price, not its 2024 launch price, and Claude 3 Opus’s is its launch price; the two starting points are not measured the same way.
The figures do not show what either company earns on a model. Prices paid across the wider market have also moved: as we reported earlier, OpenRouter said closed-weight tokens on its own traffic cost 57% less than in January.
This Week In The AI Supercycle
The Great Convergence of Capital, Power, and Geopolitics
This week’s issue: for two years the AI build-out was priced as a supply story, and last week as a credit story. This week it was priced as a permission story, and the answer arrived as a cost, not a ban.
Since March 2024, Anthropic’s Opus line got 73% cheaper per output token at list while scoring 42.7 points higher on Epoch’s index, by our arithmetic. OpenAI kept a $10 model near the frontier with GPT-6.1 Sol and priced GPT-6 Astra at $50, as Anthropic priced Claude Fable 5.1, with Epoch’s credible intervals for the top three models overlapping.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Scores are from Epoch AI’s Epoch Capabilities Index data file, downloaded on November 8, 2026. Prices are standard list prices per million tokens from OpenAI’s and Anthropic’s pricing and model pages, read the same day; Claude 3 Opus’s is from Anthropic’s March 2024 announcement. Changes and ratios are our arithmetic. This piece is not investment advice.
Gennaro Cuofano is a CRO and tech executive who has worked in AI since late 2015, starting with bringing NLP, natural-language generation, voice and chatbot products to the marketing industry. His background is in law and finance: he holds a Master's degree in Law and an International MBA with an emphasis on corporate finance (LUISS Business School and the University of San Diego, 2012), and worked as a financial analyst at a real-estate investment firm in San Diego and as an assistant controller. His work focuses on business model strategy, business engineering and, more broadly, structural analysis: how companies make money, read from their own filings. He created FourWeekMBA and leads research there and at The Business Engineer, his newsletter on AI and business strategy, with over 95,000 subscribers and more than 1,000 published analyses. His writing has also appeared on Entrepreneur (for example, How to Digitize a Family Business in Europe) and HackerNoon. Find him on LinkedIn and Substack. See how we source.
