OpenAI vs Anthropic vs Gemini: Pricing Comparison 2026

AI provider pricing has changed enough times in the past year that a comparison from six months ago is close to useless today. Here's where things stood as of early August 2026, across the three providers most teams are choosing between — with the caveat that all three companies update these rates regularly, so treat this as a snapshot, not a permanent reference.

Flagship-tier models

Mid-tier / workhorse models

*OpenAI cut Terra's price roughly 20% on July 30, 2026 — verify current rate before budgeting. **Anthropic's Sonnet 5 rate is introductory through August 31, 2026; standard pricing of $3.00/$15.00 begins September 1, 2026.

Budget / high-volume models

What actually differs beyond the headline rate

• Caching discounts: Anthropic and Google both cut cached input cost by up to 90%. OpenAI applies a smaller automatic discount (roughly 10% of the input rate) to repeated prompt prefixes.
• Batch pricing: all three offer a flat 50% discount for asynchronous, non-real-time processing — worth using for anything that doesn't need an instant response.
• Context window: most current flagship and mid-tier models across all three providers now support a 1M-token context window, though Google's Pro tier applies a higher rate above roughly 200K tokens in a single prompt.
• Output-to-input ratio: output tokens run 5-6x the input rate on every provider here, which matters more for chat-heavy or long-form generation workloads than the headline input price does.

How to actually use this table

The cheapest model on paper isn't automatically the right one — a budget-tier model that needs three retries to get a task right can end up costing more than a flagship model that gets it right the first time. The more useful exercise is matching task complexity to tier: budget/mini-class models for classification, extraction, and routing; mid-tier for most production chat and content generation; flagship reserved for tasks that genuinely need the strongest reasoning available.

Because these rates shift every few weeks — and because most teams run workloads across more than one provider — the practical answer isn't memorizing a pricing table. It's having a dashboard that pulls current usage against current rates automatically, so the comparison above stays accurate without you having to re-check it every time a provider announces a change.