Claude vs OpenAI API Cost: Four Workloads Priced From the Rate Cards

The rate cards line up more than the comparisons suggest

Most Claude versus OpenAI cost comparisons still quote older models. The current pages tell a different story. Anthropic lists Claude Sonnet 5.5 at $2 per million input tokens and $10 per million output tokens. OpenAI lists gpt-6.1-sol at the same $2.00 and $10.00 for short-context standard requests. At the small end, Claude Haiku 5.5 is $0.10 and $0.50 for prompts up to 100,000 tokens, and gpt-6-luna is $0.10 and $0.50. At the top, Claude Fable 5.1 is $10 and $50, and gpt-6-astra is $10.00 and $50.00.

One more pair matches for now. Claude Opus 5.5 is $4 and $20, and gpt-5.6-sol is $4.00 and $20.00, but OpenAI labels that gpt-5.6-sol rate promotional and says it is available at least through November 21, 2026. Treat that pair as a snapshot rather than a standing price.

Since the sticker prices match, a comparison that stops at the first two columns will conclude the providers cost the same. The rest of this article prices four hypothetical workloads to show where the totals split anyway.

Where the bills diverge: cache reads, long prompts, token counts

Cache reads. Claude Sonnet 5.5 and gpt-6.1-sol both read cached input at $0.10 per million tokens. The gap opens above that tier. Anthropic prices cache hits on Claude Opus 5.5 at $0.20 and on Claude Fable 5.1 at $0.25. OpenAI lists cached input at $0.40 for gpt-5.6-sol and $1.00 for gpt-6-astra. A workload that re-reads a large prefix on every request feels that difference.

Long prompts. Anthropic says Claude 4.6 and later models, except Claude Haiku 5.5, include the full 1M token context window at standard pricing, so a 900k-token request is billed at the same per-token rate as a 9k-token request. OpenAI splits its rates at 272K input tokens. For gpt-6.1-sol the long-context rates are $4.00 input and $15.00 output, against $2.00 and $10.00 below the threshold. The small tiers run the other way: Claude Haiku 5.5 moves to $0.50 and $2.50 above 100,000 prompt tokens, while gpt-6-luna moves to $0.20 and $0.75 above 272K.

Token counts. Anthropic states that Claude 4.7 and later models use a newer tokenizer that produces approximately 30% more tokens for the same text, with the exact increase depending on content. A per-token price is only comparable if the token counts are. OpenAI's pricing page makes no equivalent statement, so the only safe method is to count a sample of your own traffic with each provider's token counting and compare the totals.

The remaining modifiers are close. Both providers discount batch requests by 50% (Claude Sonnet 5.5 batch is $1 and $5, gpt-6.1-sol batch is $1.00 and $5.00). Anthropic charges 1.1x for US-only inference with inference_geo, and OpenAI adds 10% for regional processing endpoints on models released on or after March 5, 2026.

Four hypothetical workloads, priced from the rate cards

The traffic below is invented to show the arithmetic. Substitute your own volumes. All prices are standard, short-context or per-tier rates from the two pricing pages as read on 2026-10-22.

Workload A, retrieval Q&A with no caching: 100,000 requests a month, 6,000 input tokens and 500 output tokens each. Claude Sonnet 5.5 and gpt-6.1-sol both cost 600M x $2 + 50M x $10 = $1,700 a month on the same token counts. If Claude counts 30% more tokens for the same text, the Claude total becomes $2,210, a $510 gap that comes entirely from tokenization.

Workload B, chat with a cached prefix: 1,000,000 requests a month, each with a 4,000-token cached prefix, 300 new input tokens and 400 output tokens, assuming every prefix read is a cache hit and ignoring cache-write cost. Sonnet 5.5 and gpt-6.1-sol both come to $400 + $600 + $4,000 = $5,000. At the next tier, Claude Opus 5.5 is $800 + $1,200 + $8,000 = $10,000, against $1,600 + $1,200 + $8,000 = $10,800 for gpt-5.6-sol. At the top tier, Claude Fable 5.1 is $1,000 + $3,000 + $20,000 = $24,000, against $4,000 + $3,000 + $20,000 = $27,000 for gpt-6-astra.

Workload C, long documents: 20,000 requests a month, each with 400,000 input tokens and 2,000 output tokens. This crosses OpenAI's 272K threshold but not Anthropic's 1M window. Claude Sonnet 5.5 costs 8B x $2 + 40M x $10 = $16,400. gpt-6.1-sol at long-context rates costs 8B x $4 + 40M x $15 = $32,600. On the small tier the result reverses: Claude Haiku 5.5 at its over-100,000-token rates costs 8B x $0.50 + 40M x $2.50 = $4,100, and gpt-6-luna at long-context rates costs 8B x $0.20 + 40M x $0.75 = $1,630.

Workload D, overnight batch classification: 5,000,000 requests, 800 input tokens and 50 output tokens each, on the small tier through each batch API. Claude Haiku 5.5 batch is $0.05 and $0.25, and gpt-6-luna batch is $0.05 and $0.25, so both cost $200 + $62.50 = $262.50, half of the $525 standard total.

• A, no cache, mid tier: equal on list price, Claude up 30% if its token count is 30% higher.
• B, cached prefix: equal at mid tier, Claude cheaper at the Opus and Fable tiers.
• C, 400K-token prompts: Claude cheaper at mid tier, OpenAI cheaper at the small tier.
• D, batch at the small tier: equal.

How to choose without trusting a headline price

Pick the model by quality on your task first, because a cheaper model that needs a second attempt costs more. Then price your actual traffic. Three inputs decide which provider is cheaper: the share of input that is a repeated prefix, the share of requests above 100,000 and 272K prompt tokens, and your measured token count for the same text on each provider.

These numbers change. Anthropic notes that Claude Sonnet 5 launched at introductory pricing and that the planned increase did not happen, and OpenAI's gpt-5.6-sol rate carries a promotional end date. Re-read both pages before committing a budget, and set a provider-side spend cap so a wrong estimate has a ceiling.