Is Claude or ChatGPT cheaper on the API in 2026?

It depends on the tier. At the top end, OpenAI's GPT-5.5 and Claude Opus 4.8 share the same $5/M input price, but Claude Opus is cheaper on output ($25/M vs $30/M). For mid-tier work, OpenAI's GPT-5.4 ($2.50/$15 per M) is slightly cheaper on input than Claude Sonnet 4.6 ($3/$15), with identical output. At the budget end, OpenAI's GPT-5.4-mini and nano undercut Claude Haiku 4.5. So OpenAI tends to win on raw price at the cheap and mid tiers; Claude Opus wins on output price at the top tier.

How much does the Claude API cost per million tokens?

As of June 2026: Claude Opus 4.8 / 4.7 / 4.6 / 4.5 cost $5 input and $25 output per million tokens. Claude Sonnet 4.6 / 4.5 cost $3 input and $15 output. Claude Haiku 4.5 costs $1 input and $5 output. The Batch API gives a 50% discount and cached reads cost 0.1x the input rate.

How much does the ChatGPT (OpenAI) API cost per million tokens?

As of June 2026 (standard pricing): GPT-5.5 costs $5 input and $30 output per million tokens; GPT-5.4 costs $2.50 input and $15 output; GPT-5.4-mini costs $0.75 input and $4.50 output; GPT-5.4-nano costs $0.20 input and $1.25 output. The model behind ChatGPT (chat-latest) is $5 input and $30 output. The Batch API offers a 50% discount and cached input is roughly 0.1x the standard input rate.

Does cheaper API pricing mean lower quality?

Not directly. Price tiers track model size and capability, not vendor quality. A budget model from either provider (GPT-5.4-nano, Claude Haiku 4.5) is fine for classification, extraction, and routing, but weaker at complex reasoning than the flagship tier (GPT-5.5, Claude Opus). Match the model tier to the task rather than always reaching for the cheapest or the most expensive.

How can I cut API costs on either platform?

Both OpenAI and Anthropic offer a 50% Batch API discount for non-time-sensitive jobs and prompt caching that drops repeated-context reads to roughly 0.1x the input price. Use the smallest model that passes your eval, cap max output tokens, and cache long system prompts or documents you reuse across calls.

ChatGPT vs Claude API cost (2026): per-token pricing compared

Short answer: Neither is universally cheaper — it depends on the tier. At the flagship level, OpenAI's GPT-5.5 and Claude Opus 4.8 match on input ($5/M) but Claude Opus is cheaper on output ($25 vs $30). At the workhorse tier, GPT-5.4 ($2.50/$15) slightly undercuts Claude Sonnet 4.6 ($3/$15) on input. At the budget tier, OpenAI's mini and nano models undercut Claude Haiku 4.5. Both offer a 50% Batch discount and ~0.1x cached input.

If you're picking an API provider on cost, the "ChatGPT vs Claude" question is really a per-tier question: each provider has a flagship, a workhorse, and one or more budget models, and the winner flips depending on which row you're buying. Below is the current per-million-token pricing for both, pulled from OpenAI's and Anthropic's official pricing pages (June 2026), followed by which one is cheaper for each common workload.

API pricing: OpenAI vs Anthropic (per 1M tokens)

Prices are standard, pay-as-you-go, in USD per million tokens. "Cached input" is the discounted rate when a request reuses previously processed context.

Tier	OpenAI (ChatGPT / GPT-5)	Input	Output	Anthropic (Claude)	Input	Output
Flagship	GPT-5.5	$5.00	$30.00	Claude Opus 4.8	$5.00	$25.00
Workhorse	GPT-5.4	$2.50	$15.00	Claude Sonnet 4.6	$3.00	$15.00
Budget	GPT-5.4-mini	$0.75	$4.50	Claude Haiku 4.5	$1.00	$5.00
Ultra-budget	GPT-5.4-nano	$0.20	$1.25	(no nano tier)	—	—
ChatGPT model	chat-latest	$5.00	$30.00	Claude Sonnet 4.6	$3.00	$15.00

Bold = cheaper of the two in that row. Both providers price a 1M-token context window at the standard rate on their long-context flagship models. Source: OpenAI and Anthropic official pricing pages, June 2026.

Which is cheaper for your workload

"I need the best reasoning quality, cost is secondary"

Flagship tier: Claude Opus 4.8 and GPT-5.5 tie on input ($5/M) but Opus is cheaper on output ($25 vs $30/M). For output-heavy generation (long drafts, large code files), Claude Opus is the cheaper flagship; for input-heavy work (stuffing big context, short answers), they're effectively level.

"Production workhorse — high volume, good-enough quality"

This is where most real spend lives. GPT-5.4 ($2.50/$15) slightly undercuts Claude Sonnet 4.6 ($3/$15) on input with identical output pricing. On a typical mixed workload they land within ~10% of each other, so pick on quality fit and latency rather than price — the gap is small enough to be noise next to how well each model does your specific task.

"High-volume classification, extraction, routing"

Budget tier: OpenAI wins outright. GPT-5.4-mini ($0.75/$4.50) and GPT-5.4-nano ($0.20/$1.25) are cheaper than Claude Haiku 4.5 ($1/$5), and nano has no direct Claude equivalent. For millions of cheap calls (tagging, moderation, simple structured output), OpenAI's budget models are the lower-cost option.

"Reusing a big system prompt or document across many calls"

Both providers offer prompt caching that drops repeated-context reads to roughly 0.1x the input price (e.g. Claude Opus cache reads at $0.50/M vs $5 standard; GPT-5.5 cached input at $0.50/M vs $5). If your prompts share a large fixed prefix, caching matters more than the headline per-token gap — and both implement it similarly.

"Non-time-sensitive batch jobs"

Both offer a 50% Batch API discount on input and output. That halves every number in the table above for asynchronous workloads, and it stacks with prompt caching on both platforms.

Costs that aren't per-token

Web search tool: both charge $10 per 1,000 searches on top of token costs for reasoning models.
Data residency / regional endpoints: OpenAI adds a 10% uplift on regional endpoints for the GPT-5.5/5.4 family; Anthropic adds a 1.1x multiplier for US-only inference on Opus 4.6 / Sonnet 4.6 and later.
Long context: OpenAI's flagship charges a higher "long context" rate above ~272K input tokens; Anthropic bills the full 1M-token window at the standard per-token rate on its long-context models.

The bottom line

If you're optimizing purely for API cost: go OpenAI at the budget and ultra-budget tiers (mini/nano undercut Haiku), call it a near-tie at the workhorse tier (GPT-5.4 vs Sonnet 4.6), and lean Claude Opus at the flagship tier if your workload is output-heavy. But the per-token gaps inside a tier are small — for most teams the bigger lever is choosing the right tier for each task and using batch + caching, not switching vendors over a 10–20% line-item difference.