Claude Haiku 4.5 vs GPT-5 nano pricing
Side-by-side LLM API pricing.GPT-5 nano is cheaper on input ($0.05 vs $1.00 /1M), and GPT-5 nano is cheaper on output ($0.40 vs $5.00 /1M). On a blended average, GPT-5 nano is lower — but which wins for you depends on your input/output mix, so check the example workloads below or run the calculator.
| Attribute | Claude Haiku 4.5 | GPT-5 nano |
|---|---|---|
| Provider | Anthropic | OpenAI |
| Tier | Budget | Budget |
| Status | GA | Legacy |
| Context window | 200K | — |
| Input /1M | $1.00 | $0.05 |
| Output /1M | $5.00 | $0.40 |
| Cached input /1M | $0.10 | $0.005 |
| Batch input /1M | $0.50 | $0.025 |
| Batch output /1M | $2.50 | $0.20 |
| Blended (avg in+out) | $3.00 | $0.225 |
| 1M in + 1M out | $6.00 | $0.4500 |
| 1M in (90% cached) + 100K out | $0.6900 | $0.0495 |
List prices, USD, directional. Rates are provider list prices per 1M tokens and are meant for comparison, not billing. Batch rates shown as 50% off are derived where a provider offers batch but does not publish a separate figure. Preview, promo, intro, peak/off-peak, long-context, and third-party-host prices are labeled where they apply. Token counts vary by tokenizer, so per-token price is not always a like-for-like cost. Always confirm with the provider before relying on a number. Prices as of 2026-08-03.
Which should you pick?
For a balanced job (1M input + 1M output), GPT-5 nano costs $0.4500 versus $6.00. If your prompts are large and reused, prompt caching changes the maths — the cache-heavy row above shows a 90%-cached input scenario. If your work can run asynchronously, both providers' batch rates cut the bill further where offered.
Full price matrices: Claude Haiku 4.5 pricing → · GPT-5 nano pricing → · Back to the LLM pricing hub →