
Which AI company is winning? A spending chart from presentation-software company Gamma offers an interesting answer to a narrower question: which suppliers are doing which jobs inside one application. Reading it as a ranking of the entire AI market would go much further than the evidence allows.
In its November 2026 Model Index, Gamma says OpenAI overtook Anthropic in its own usage after a set of late-October tests. Its companion article explains how smaller models can handle much of the work involved in creating presentations. Both are first-party accounts of Gamma’s product, not independent industry surveys.
Start with the period and the denominator
The index’s written account puts OpenAI at 40% of Gamma’s AI spending in the last full week of October, followed by Anthropic at 38% and Google at 9%. Those figures describe that specific period and Gamma’s spending. They do not measure global AI revenue, the number of businesses using each provider, or general model quality.
Gamma also says it values usage at list prices. That matters: a list-price comparison is not necessarily the same thing as a company’s final invoice after commercial discounts. Its figures include both language and image models in the relevant spending comparisons, so a language-only comparison would answer a different question.
A high call count can still mean a smaller bill
For the week covered by its volume discussion, Gamma says its software made more than 70 million model calls. Three small models together handled about three-quarters of the calls for roughly one-quarter of language-model spending. These are Gamma’s reported operating figures, not results independently audited by BizTechIndia.
The distinction is simple. Imagine one supplier performs many inexpensive checks while another handles fewer, more costly tasks. The first could lead in call volume and the second in spending. Neither metric alone tells a buyer which system will work best for a different task.
Task results matter more than a single winner
Gamma reports using different models for different parts of its product. It evaluates outcomes such as whether people keep edits, finish presentations, share or export their work, as well as speed and cost. Its conclusion is that a model can perform well on one job and poorly on another.
That is a useful purchasing lesson, but not a universal benchmark result. A sales team summarising documents, a developer writing code and a designer repairing slides have different success criteria. Gamma’s evidence is most directly relevant to its own presentation-making workload.
Lower cost does not erase the quality trade-off
The companion article acknowledges that about 4% of slides in its smaller-model configuration still failed a check after repair, compared with under 2% when larger models performed the same jobs. This is a measure of Gamma’s checks, not a claim that the remaining slides are free of factual or visual errors.
For a business considering AI software, the practical question is therefore not simply the price of a model call. It is the cost of reaching an acceptable finished result, including review and correction. A cheaper first attempt may be worthwhile, but only if the remaining error rate fits the job.
How to read the next AI usage headline
- Check whose usage is being measured: one application or an industry-wide sample.
- Keep the reporting period attached to every percentage.
- Separate calls, tokens, spending and final business outcomes.
- Ask whether prices are list prices or actual paid costs.
- Look for the tasks, evaluation method and remaining failure rate behind a performance claim.
Gamma’s reports provide a useful case study of an application choosing between suppliers. They do not settle the AI race. The more useful takeaway is a habit: define the job and the measurement before declaring a winner.
Sources and scope
- Gamma Research: The Gamma Model Index, November 2026.
- Gamma Research: Getting frontier performance without frontier models, November 2026.
Both sources were checked on November 17, 2026. Figures and product-test findings are attributed to Gamma; BizTechIndia has not independently reproduced the experiments. Original explanation and illustration by BizTechIndia. The artwork is illustrative, not a plotted data chart.