GPT vs. Claude vs. Gemini: API Pricing Compared
Input and output tokens are priced separately by every provider, and the gap between the two matters more than the headline rate — a model that looks expensive on input can still be the cheapest choice for an output-light workload. These are the current standard-tier rates for each provider's mainstream model, per million tokens.
| Criterion | GPT-5.2 Standard | Claude 4.6 Sonnet | Gemini 3.1 Pro |
|---|---|---|---|
| Provider | OpenAI | Anthropic | |
| Input cost (per 1M tokens) | $1.75 | $3.00 | $2.00 |
| Output cost (per 1M tokens) | $14.00 | $15.00 | $12.00 |
| Output-to-input price ratio | 8.0x | 5.0x | 6.0x |
Output cost (per 1M tokens): Output is always the larger share of the bill for most workloads — weight this row more than input cost when comparing.
Output-to-input price ratio: Lower is better for output-heavy workloads (long-form generation, code). Claude 4.6 Sonnet has the smallest output premium of the three.
Frequently Asked Questions
Which is cheapest for a typical chatbot workload?
For a short-input, short-output workload, GPT-5.2 Standard has the lowest input cost and a competitive output cost, making it the cheapest of the three for high-volume, low-length requests.
Which is cheapest for long-form generation?
Gemini 3.1 Pro has the lowest output cost per token ($12.00/1M), which matters most for output-heavy workloads like long-form content generation or code generation where output volume dominates the bill.
Does a lower output-to-input ratio always mean cheaper overall?
No — it means the output premium is smaller relative to that model's own input price, not that the model is cheaper in absolute terms. Claude 4.6 Sonnet's ratio is lowest, but its absolute input and output prices are both higher than GPT-5.2 Standard's. Use the AI Token Calculator to compare exact costs for your actual prompt volume.