GPT-5.4 pricing

OpenAI · API id gpt-5.4 · verified 2026-11-17

Input$2.50
Cached input (cache hit)$0.25
Cache write—
Output$15.00
Batch discount50%
Context window1,050,000 tokens
Max output128,000 tokens
Above 272K prompt tokens$5.00 in / $22.50 out

USD per 1M tokens unless noted. Official source

Estimate your workload on GPT-5.4 →

What GPT-5.4 costs for real workloads

WorkloadGPT-5.4Mistral Large 3Mistral Large 4DeepSeek V4 ProGrok 4.7
10,000 chatbot replies
2,000 input + 400 output tokens each, 50% of the prompt cached
$87.50$11.50$16.10$29.48$49.00
One 30-step coding agent run
12K system + tools prompt, 1K output and 3K tool result per step, caching on
$1.28$0.21$0.30$0.38$1.45
1M documents classified (batch)
600 input + 20 output tokens each, batch API where available
$900$165$231$871$1,056

Compared with the cheapest other flagship models. Change the numbers in the calculator.

Billing notes

Older generation (snapshot gpt-5.4-2026-03-05). No cache-write price listed. Long context: prompts >272K input tokens bill the whole request at 2x input/cache and 1.5x output. Reasoning tokens billed as output. Regional/data-residency endpoints +10%.

Where GPT-5.4 ranks by use case

Position by cost among all models that fit each workload (1 = cheapest).

Counting tokens for a prompt? The OpenAI token counter counts GPT-5.4 tokens (OpenAI (o200k_base) tokenizer) and shows the input and output cost.

Head-to-head comparisons

Other OpenAI models

All 45 models side by side: LLM API pricing comparison.

FAQ

How much does GPT-5.4 cost per 1M tokens?

GPT-5.4 costs $2.50 per 1M input tokens and $15.00 per 1M output tokens at standard rates (verified 2026-11-17).

Does GPT-5.4 support prompt caching?

Yes. Cached input is billed at $0.25 per 1M tokens.

Is there a batch discount for GPT-5.4?

Yes. The batch API is 50% cheaper than standard requests.

What is the context window of GPT-5.4?

1,050,000 tokens, with up to 128,000 output tokens. Prompts over 272,000 tokens are billed at $5.00 input and $22.50 output for the whole request.