AI Rate CardClaude Haiku 4.5 (latest) vs GPT-5.3 Codex: API price and specs
Claude Haiku 4.5 (latest) is 64% cheaper than GPT-5.3 Codex on input price.
Claude Haiku 4.5 (latest) is 78% cheaper than GPT-5.3 Codex on output price.
Claude Haiku 4.5 (latest) is 73% cheaper than GPT-5.3 Codex on a 1M-in / 200k-out workload.
GPT-5.3 Codex has a 100% larger context than Claude Haiku 4.5 (latest).
| Field | Claude Haiku 4.5 (latest) | GPT-5.3 Codex |
|---|---|---|
| Lowest input, per 1M | $0.50 | $1.40 +180% |
| Median input, per 1M | $1.00 | $1.75 +75% |
| Lowest output, per 1M | $2.50 | $11.20 +348% |
| Median output, per 1M | $5.00 | $14.00 +180% |
| Lowest unit price | $0.001 / request | — |
| Cost of 1M in + 200k out | $1.00 | $3.64 +264% |
| Context | 200K +50% | 400K |
| Max output | 64K +50% | 128K |
| Providers | 34 | 22 +35.3% |
| Regions | APAC, AU, EU, GLOBAL, IN, JP, US | GLOBAL, US |
| Accepts | application/pdf, file, image, image/gif, image/jpeg, image/png, image/webp, pdf, text | application/pdf, file, image, image/gif, image/jpeg, image/png, image/webp, pdf, text |
| Returns | text | text |
| Type | Text models | Text models |
| Free option | no | no |
| Cheapest provider | openrouter | cometapi |
Your own workload: how many tokens the model gets (input) and gives back (output) - for example, in a month. The "Cost" row of the table above is recalculated for these numbers, for each model.
Computed from current offers each time the page opens; the best value of a row is bold, the percent is how far behind the best.
Providers selling both
abacus, aihubmix, aiml, azure, cometapi, digitalocean, edenai, fastrouter, freemodel, github-copilot, impossibl, jiekou, kilo, ofox, opencode, openrouter, orcarouter, poe, requesty, vercel, zenmux
Frequently asked questions
Which is cheaper, Claude Haiku 4.5 (latest) or GPT-5.3 Codex?
Claude Haiku 4.5 (latest) is 64% cheaper than GPT-5.3 Codex on input price. Claude Haiku 4.5 (latest) is 78% cheaper than GPT-5.3 Codex on output price. Claude Haiku 4.5 (latest) is 73% cheaper than GPT-5.3 Codex on a 1M-in / 200k-out workload.
Which has the larger context, Claude Haiku 4.5 (latest) or GPT-5.3 Codex?
Claude Haiku 4.5 (latest): 200K tokens; GPT-5.3 Codex: 400K tokens.
Where can I use both Claude Haiku 4.5 (latest) and GPT-5.3 Codex?
abacus, aihubmix, aiml, azure, cometapi, digitalocean, edenai, fastrouter, freemodel, github-copilot, impossibl, jiekou, kilo, ofox, opencode, openrouter, orcarouter, poe, requesty, vercel, zenmux