OpenAI had 81.1% of low-end tokens on proprietary AI models routed through Databricks’ Unity Gateway in the week of 2 November, venture investor Tomasz Tunguz wrote on 16 November, reading the Databricks AI Index. Anthropic had 94.1% of the top, state-of-the-art tier.
Priced at list, the split is less even. Our number: OpenAI’s low-end models took 40.3% of that tier’s spend, and Anthropic’s top tier held 45.4% of all proprietary-model spend against 7.7% for OpenAI’s low end, about 5.9 times, by our arithmetic from the index’s model-level values.
Business Pill · Usage-Based Pricing
An evergreen Business Engineer explainer on usage-based pricing, paying for work rather than seats. It teaches the general idea only and does not discuss Databricks, Anthropic or OpenAI.
The key insight: As we read the Databricks AI Index with Tomasz Tunguz’s tiers, the two labs lead tiers of very different dollar size: OpenAI had 81.1% of low-end tokens but 40.3% of low-end list-price spend in the week of 2 November, while Anthropic’s top-tier spend alone was 45.4% of all proprietary spend, by our arithmetic.
What Tunguz Wrote
Tunguz wrote on X that “OpenAI & Anthropic each hold roughly 90% market share of an AI segment, but those segments are different”. In the full post he puts Anthropic at 94% of the premium end and OpenAI at 81% of the low end, across what he describes as 10,000+ enterprises.
He asks whether the split is “a deliberate pricing strategy to capture different ends of the market, or a structural cost difference in how the two labs serve tokens?” His data is the Databricks AI Index, and his footnote says Unity Gateway “reflects Databricks’ enterprise customer base (data-platform & agentic-coding workloads) rather than the entire enterprise AI market”.
Both segments sit inside enterprise API traffic and are split by model tier, not consumer against business. His top tier counts Anthropic’s Opus and Fable models against OpenAI’s GPT 6 Astra and GPT 5.6 Sol. His low end counts GPT 6 Luna, GPT 5 Nano and older OpenAI models against older Anthropic models, a bucket that includes Haiku.

His Token Figures Check Out
We re-summed his tiers from the index’s model-level token values and got the same results for the week of 2 November: Anthropic 94.1% of the top tier, OpenAI 46.9% of the mid tier and OpenAI 81.1% of the low end. In the week of 20 July, the first the index plots, Anthropic had 81.3% of the top tier and OpenAI 30.3% of the low end, by our arithmetic.
Tunguz wrote that the mid tier is where the two labs now meet: OpenAI went from 0% of mid-tier volume in July to 47% by 2 November, comparing Sonnet 5 and Sonnet 5.5 with GPT-6 Sol and GPT-6.1 Sol.
In Dollars, the Low End Splits Differently
The index also prices every token at each provider’s list price, and the page says that is “not what Databricks bills”. On that measure, with Tunguz’s tiers, OpenAI’s low-end models took 40.3% of low-end spend in the week of 2 November and older Anthropic models 59.7%, by our arithmetic. Anthropic’s top-tier share moves less, from 94.1% of tokens to 82.7% of spend.
The gap comes from the bucket. Older Anthropic models drew about 2.5 times the average proprietary spend per token, and OpenAI’s low-end models about 0.40 times, by our arithmetic. Tunguz notes that the index does not report Haiku on its own, so the figures do not show which Anthropic models carry that spend.

Where the Dollars Sit
The tiers differ in size. The top tier held 42.3% of proprietary tokens and 54.9% of list-price spend in the week of 2 November; the low end held 23.8% of tokens and 19.1% of spend, by our sums. Anthropic’s top-tier spend alone came to 45.4% of all proprietary spend on the gateway.
Our earlier piece on the index found Anthropic with 70.1% of closed-model spend and 61.5% of tokens in the same week. An OpenRouter read in October found Anthropic drawing 2.1 times OpenAI’s spend per token.
What the Price Lists Show
On the two labs’ pricing pages, read on 17 November, Anthropic’s most-used top-tier model on the gateway lists below OpenAI’s: Claude Opus 5.5 is $4 per million input tokens and $20 per million output, and GPT-6 Astra is $10 and $50. Opus 5.5 is 40% of Astra’s list price, by our arithmetic. Fable 5.2 lists at $8 and $40.
Lower down, the lists match. Claude Sonnet 5.5 and GPT-6.1 Sol both list at $2 and $10. Claude Haiku 5.5 and GPT-6 Luna both list at $0.10 and $0.50, Haiku’s price for prompts up to 100,000 tokens.
OpenAI’s Luna cuts were steep: input went from $1.00 to $0.20 per million tokens on 30 July, as Tunguz lists them, and to $0.10 with GPT-6 Luna, 90% in all. OpenAI’s September post gives a cost reason: “Improvements in caching and inference let us serve these models at lower cost, and we’re passing those savings directly on to users and customers”.
Other Data Splits the Labs Differently
Other customer data cuts the market another way. In the Ramp AI Index, Anthropic took 49.3% of token spending in the week ending 1 November and OpenAI 46.1%, and at least 28.6% of all Ramp businesses paid both labs in October, by our arithmetic in that piece.
Gamma, a single heavy buyer, put OpenAI at 40% of its AI spend and Anthropic at 38% in the last full week of October, after GPT-6.1 Terra replaced Claude Haiku 5.5 and Sonnet 5.5 in several jobs.
None of these counts consumers. OpenAI CFO Sarah Friar said on stage that “1.2 billion people come to ChatGPT every single week”; Tunguz’s split covers enterprise API tokens on one gateway.
The Structural Read
As we read the price lists, Anthropic’s top-tier lead does not come with a higher list price: Opus 5.5 lists at $4 and $20 per million input and output tokens against $10 and $50 for GPT-6 Astra, while Sonnet 5.5 and GPT-6.1 Sol, and Haiku 5.5 and GPT-6 Luna, list at the same prices.
As we read OpenAI’s own posts, the company describes its Luna cuts as cost savings passed on, but neither lab publishes serving costs, so the figures here do not settle which of Tunguz’s two explanations holds.
Tomasz Tunguz, on tomtunguz.com
“is Luna’s pricing a short-term market-share subsidy, or a structural cost advantage in OpenAI’s model architecture?”
Three Implications
FOR BUYERS Tier shares in tokens and in dollars differ by up to 40.8 points, by our arithmetic, so the count used changes which lab looks larger.
FOR MODEL MAKERS The top tier held 54.9% of list-price spend on proprietary models in the week of 2 November, against 19.1% for the low end, by our sums.
RELATED Our first piece on the Databricks AI Index gives the provider totals.
The Business Engineer Lens
This story maps onto the Business Engineer framework The CFO’s Guide to the Token Economy.
The framework argues: “As companies move from chatbots to thousands of autonomous agentic workflows running 24/7, the line item that used to be a rounding error becomes the budget.”
As we read it, the tier split shows where that budget goes on one gateway: the low end, where OpenAI has 81.1% of tokens after its Luna cuts, held 19.1% of list-price spend, while the top tier, where Anthropic leads, held 54.9%. The framework predates Tunguz’s post and does not discuss it.
What Is Not Established
The index does not say why each lab leads its tier, and neither pricing page gives serving costs, so these figures do not settle Tunguz’s question of pricing strategy or structural cost. Spend shares are at list price and cover Unity Gateway customers only; the tier totals are our sums using Tunguz’s tier definitions.
We read Tunguz’s post in full, the index page including its chart values, both labs’ pricing pages and OpenAI’s two pricing posts. We did not contact Tunguz, Databricks, Anthropic or OpenAI for this piece.
The Bottom Line
Tunguz’s token split holds in the Databricks AI Index: Anthropic 94.1% of the top tier, OpenAI 81.1% of the low end. At list price, OpenAI’s low-end share is 40.3%, and Anthropic’s top tier holds 45.4% of all proprietary spend, by our arithmetic.
95,000+ executives read Business Engineer for the AI strategy frameworks cited by ChatGPT, Claude, and Perplexity.
A note on sourcing. Tomasz Tunguz’s figures and question come from his post of 16 November 2026 on tomtunguz.com, read in full with its footnotes, and his X post the same evening. Token and spend shares come from the Databricks AI Index page, including the exact values its charts plot (updated 2 November 2026); spend is at providers’ list prices, not Databricks’ billing. Tier totals, the dollar shares and the multiples are our arithmetic from those values, using Tunguz’s tier definitions.
List prices are from OpenAI’s and Anthropic’s pricing pages, read on 17 November 2026. Tunguz, Databricks, Anthropic and OpenAI were not contacted for this piece.
Sources: Tomasz Tunguz, Strategic Decision or Structural Advantage?, 16 Nov 2026 · Tomasz Tunguz on X, 16 Nov 2026 · Databricks AI Index (page updated 2 Nov 2026) · OpenAI API pricing page (read 17 Nov 2026) · Anthropic pricing page (read 17 Nov 2026) · OpenAI, Introducing GPT-6 Sol and Luna, 22 Sep 2026 · OpenAI, GPT-5.6 price-performance post, 30 Jul 2026









