Anthropic vs OpenAI API Pricing 2026: Per-Token Cost Compared

Through mid-2026 the two price lists differed only slightly at most tiers. The September and October 2026 releases from both companies closed even that gap: Claude Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5 on one side, GPT-6 Astra, GPT-6.1 Sol and GPT-6 Luna on the other. This article lays the current rates side by side, shows where the two lists still differ, and runs a SaaS margin scenario through the calculator so the cost shows up as gross margin rather than abstract per-token rates.

1. The per-token rate table

Standard (non-batch) rates per million tokens, read from both official pricing pages on October 18, 2026.

Tier	Anthropic model	Input / Output	Cache hit	OpenAI model	Input / Output	Cached input
Flagship	Claude Fable 5.1[1]	$10 / $50	$0.25	GPT-6 Astra[2]	$10 / $50	$1.00
Second flagship	Claude Opus 5.5[1]	$4 / $20	$0.20	GPT-5.5[2]	$5 / $30	$0.50
Workhorse	Claude Sonnet 5.5[1]	$2 / $10	$0.10	GPT-6.1 Sol[2]	$2 / $10	$0.10
Small / fast	Claude Haiku 5.5[1]	$0.10 / $0.50	$0.01	GPT-6 Luna[2]	$0.10 / $0.50	$0.01

Haiku 5.5's rate applies to prompts up to 100,000 tokens and OpenAI's rates to prompts up to 272,000 input tokens; section 2 covers what happens above those lines. Older models such as Claude Sonnet 4.6 ($3 / $15) and GPT-5.4 ($2.50 / $15) are still served, at their old prices, so moving a workload to the current generation is itself a price cut.

2. Where the two price lists differ

With the headline rates matched, three details decide which bill is lower:

• Cached input at the top tier. Claude Fable 5.1 bills a cache hit at $0.25 per million tokens, 2.5% of its input rate. GPT-6 Astra bills $1.00, 10%. An agent that resends a long system prompt or document on every call pays up to 4x less for that context on Fable 5.1. At the workhorse and small tiers the cached rates are identical.
• Long prompts. OpenAI bills any request with more than 272,000 input tokens at its long-context rate for the whole request: GPT-6 Astra $20 / $75, GPT-6.1 Sol $4 / $15, GPT-6 Luna $0.20 / $0.75^[2]. Anthropic bills Fable 5.1, Opus 5.5 and Sonnet 5.5 at the standard rate across the full 1M-token window; Haiku 5.5 is the exception, at $0.50 / $2.50 for prompts over 100,000 tokens^[1].
• The second flagship. Claude Opus 5.5 at $4 / $20 is 20% cheaper on input and a third cheaper on output than GPT-5.5 at $5 / $30. For heavy reasoning work that does not need the top model, this is the one tier where list price favours one vendor outright.

Both vendors add 10% for in-region processing on recent models: Anthropic's US-only inference option multiplies every token price by 1.1, and OpenAI charges a 10% uplift on data-residency endpoints for models released since March 5, 2026^[1][2].

3. Prompt caching changes the effective rate

On Anthropic, a 5-minute cache write costs 1.25x the input rate and a 1-hour write 2x; a cache hit then costs 0.1x on most models, 0.05x on Opus 5.5 and Sonnet 5.5, and 0.025x on Fable 5.1, so caching pays off after the first read on the short cache^[1]. OpenAI lists cached-input and cache-write prices per model: $1.00 and $12.50 per million on GPT-6 Astra, $0.10 and $2.50 on GPT-6.1 Sol^[2].

For an agent that resends a large system prompt, a long document, or conversation history on every call, the cached portion dominates input volume. If 80% of the input tokens are cache hits, the effective input rate on Sonnet 5.5 or GPT-6.1 Sol drops from $2 to about $0.48 per million (0.8 × $0.10 + 0.2 × $2), before cache-write costs. On the flagships the same 80% gives about $2.20 on Fable 5.1 and $2.80 on GPT-6 Astra.

4. Margin impact on a $29 SaaS

The margin calculator below prices a $29/month SaaS where each user triggers 20 API calls a day at 500 input and 1,000 output tokens, first at the workhorse rate both vendors now charge ($2 / $10: Claude Sonnet 5.5 or GPT-6.1 Sol), then at the small-model rate ($0.10 / $0.50: Claude Haiku 5.5 or GPT-6 Luna). The engine renders the gross margin from those rates:

Because the two vendors charge the same rate at each of these tiers, switching vendors changes nothing in this scenario; switching tiers changes the API line twenty-fold. Output tokens dominate the bill here (1,000 output against 500 input per call, at five times the input rate), so shorter completions and moving routine calls to the small tier move margin far more than the choice between Anthropic and OpenAI. The cheapest LLM API ranking covers the full provider field.

5. Decision guidance

• Pick the tier first, on capability. At matched tiers the per-token prices are the same, so the model that does your task better is also the cheaper one per successful result.
• Large repeated context at the top tier: Claude Fable 5.1, whose cache hits cost a quarter of GPT-6 Astra's.
• Prompts over 272,000 tokens: Anthropic's Fable 5.1, Opus 5.5 or Sonnet 5.5, billed at the standard rate across the full 1M window, where OpenAI's long-context rate doubles input.
• Heavy reasoning below the flagship: Claude Opus 5.5 at $4 / $20 against GPT-5.5 at $5 / $30.
• High-volume classification: Haiku 5.5 or GPT-6 Luna at $0.10 / $0.50; keep Haiku prompts under 100,000 tokens or its rate steps up.

Re-check both pricing pages before committing a model to production; rates at this layer move with every release. For the broader field including open-weight and low-cost providers, see the cheapest LLM API ranking and DeepSeek vs Gemini pricing.

All per-token figures read from the official Anthropic and OpenAI pricing pages on 2026-10-18.

Frequently asked questions

Is Claude or GPT cheaper per token in 2026?

At matched tiers the standard rates are now identical, checked on both official pricing pages on October 18, 2026: Claude Fable 5.1 and GPT-6 Astra are both $10 input / $50 output per million tokens, Claude Sonnet 5.5 and GPT-6.1 Sol are both $2 / $10, and Claude Haiku 5.5 and GPT-6 Luna are both $0.10 / $0.50. The one tier where they differ is the step below the flagship: Claude Opus 5.5 is $4 / $20 against GPT-5.5 at $5 / $30.

How much does prompt caching reduce cost?

Both vendors bill a cache hit at a small fraction of the input rate, but not the same fraction at the top. Claude Fable 5.1 charges $0.25 per million cached tokens against GPT-6 Astra's $1.00, a 4x gap. At the mid and small tiers the cached rates match: $0.10 on Sonnet 5.5 and GPT-6.1 Sol, $0.01 on Haiku 5.5 and GPT-6 Luna. Anthropic also charges for cache writes ($12.50 per million on Fable 5.1 for a 5-minute cache), and OpenAI now lists cache-write prices too.

Do long prompts cost more?

On OpenAI, a request with more than 272,000 input tokens is billed at the long-context rate for the whole request: GPT-6 Astra $20 / $75, GPT-6.1 Sol $4 / $15, GPT-6 Luna $0.20 / $0.75. Anthropic bills Fable 5.1, Opus 5.5 and Sonnet 5.5 at the standard rate across the full 1M-token window; Haiku 5.5 steps up to $0.50 / $2.50 for prompts over 100,000 tokens.

Which has the better batch discount?

Anthropic's Batch API is 50% off both input and output tokens on every current Claude model. OpenAI prices Batch and Flex as separate tiers on its pricing page. For non-urgent jobs such as bulk classification or document processing, the batch path is the cheapest route on both.