No-BS Take · Pricing
Claude vs OpenAI API Pricing: The Same Price at Every Tier, to the Cent
The numbers
| OpenAI (per 1M tokens, up to 272K input)Official price | GPT-6 Astra $10.00 / $50.00; GPT-6.1 Sol $2.00 / $10.00; GPT-6.1 Terra $1.00 / $5.00; GPT-6 Luna $0.10 / $0.50 | OpenAI API pricingChecked Nov 8, 2026 by AISelect editorial research, high confidence |
| Anthropic (per 1M tokens)Official price | Claude Fable 5.1 $10 / $50; Opus 5.5 $4 / $20; Sonnet 5.5 $2 / $10; Haiku 5.5 $0.10 / $0.50 (prompts up to 100K tokens) | Claude pricingChecked Nov 8, 2026 by AISelect editorial research, high confidence |
| Google (per 1M tokens)Official price | Gemini 4 Argon $2.00 / $10.00 through Nov 30, $4.00 / $20.00 from Dec 1, 2026; Gemini 3.8 Flash $0.75 / $3.75, $1.50 / $7.50 from Jan 1, 2027 | Gemini API pricingChecked Nov 8, 2026 by AISelect editorial research, high confidence |
| Claude Sonnet 5.5 and GPT-6.1 Sol at the clouds (per 1M tokens)Tracked list price | Sonnet 5.5 $2 / $10 at AWS Bedrock, Azure and Google Vertex AI; GPT-6.1 Sol $2 / $10 at AWS Bedrock and Azure | AWS Bedrock pricingChecked Nov 8, 2026 by AISelect daily price tracking (LiteLLM, models.dev), medium confidence |
| GPQA Diamond (Epoch AI, reasoning setting in brackets)Third-party benchmark | Gemini 4 Argon 96.4% (high); GPT-6 Astra 95.8% (max); Claude Sonnet 5.5 95.6% (max); GPT-6.1 Sol 95.4% (max); Gemini 3.8 Flash 95.4% (high); Claude Opus 5.5 90.6% (max); GPT-6 Luna 90.5% (max); Claude Haiku 5.5 89.6% (max) | Epoch AI benchmarks (CC BY 4.0)Checked Nov 8, 2026 by AISelect daily benchmark tracking (Epoch AI), medium confidence |
| Monthly cost for 10M input + 2M output tokensOur calculation | $10 / $50 step $200; $4 / $20 step $80; $2 / $10 step $40; Gemini 3.8 Flash $15; $0.10 / $0.50 step $2 | Our calculation from the listed pricesChecked Nov 8, 2026 by AISelect editorial research, high confidence |
Frozen at publication. Current prices are shown below.
Put OpenAI's and Anthropic's API price lists side by side and they line up exactly. GPT-6 Astra and Claude Fable 5.1 both cost $10 / $50 per million input / output tokens. GPT-6.1 Sol and Claude Sonnet 5.5 both cost $2 / $10. GPT-6 Luna and Claude Haiku 5.5 both cost $0.10 / $0.50. Not a cent apart, and every one of them bills output at exactly five times input. On December 1 Google steps onto the same ladder: Gemini 4 Argon moves from Sonnet's $2 / $10 to Claude Opus 5.5's $4 / $20.
So at list price, switching between the big labs saves you nothing. The money is in which step you pick.
The ladder
| Per 1M tokens | OpenAI | Anthropic | |
|---|---|---|---|
| $10 / $50 | GPT-6 Astra | Claude Fable 5.1 | |
| $4 / $20 | Claude Opus 5.5 | Gemini 4 Argon (Dec 1) | |
| $2 / $10 | GPT-6.1 Sol | Claude Sonnet 5.5 | Gemini 4 Argon (to Nov 30) |
| $1 / $5 | GPT-6.1 Terra | ||
| $0.10 / $0.50 | GPT-6 Luna | Claude Haiku 5.5* |
Input / output prices, standard API, from the official OpenAI, Anthropic and Google pricing pages, checked on November 8, 2026. OpenAI's prices apply up to 272K input tokens. *Haiku 5.5 for prompts up to 100K tokens.
For one month of 10 million input and 2 million output tokens, the five steps cost $200, $80, $40, $20 and $2 (our arithmetic, without caching or batch discounts).
Google's Gemini 3.8 Flash is the one big model off the ladder: $0.75 / $3.75, also five to one, and it doubles to $1.50 / $7.50 on January 1, 2027.
The step tells you the price, not the score
GPQA Diamond scores from Epoch AI, at the reasoning setting in brackets:
| Step | Model | GPQA Diamond |
|---|---|---|
| $10 / $50 | GPT-6 Astra | 95.8% (max) |
| $4 / $20 | Gemini 4 Argon (from Dec 1) | 96.4% (high) |
| $4 / $20 | Claude Opus 5.5 | 90.6% (max) |
| $2 / $10 | Claude Sonnet 5.5 | 95.6% (max) |
| $2 / $10 | GPT-6.1 Sol | 95.4% (max) |
| $0.75 / $3.75 | Gemini 3.8 Flash | 95.4% (high) |
| $0.10 / $0.50 | GPT-6 Luna | 90.5% (max) |
| $0.10 / $0.50 | Claude Haiku 5.5 | 89.6% (max) |
Epoch AI has no GPQA Diamond run for Claude Fable 5.1 yet.
Two things stand out. Claude Opus 5.5 costs twice as much as Sonnet 5.5 and scores five points lower on this test. And Gemini 3.8 Flash scores within a fraction of a point of Sonnet 5.5 and GPT-6.1 Sol for less than 40% of their price. GPQA is one set of 198 graduate-level science questions; it says nothing about long agentic coding sessions, where Opus may well earn its price. But it is reason enough to test before you pay for the next step up.
Where the prices do differ
- Long prompts. OpenAI charges GPT-6.1 Sol $4 / $15 and GPT-6 Luna $0.20 / $0.75 above 272K input tokens. Anthropic keeps Sonnet 5.5 on one rate but charges Haiku 5.5 $0.50 / $2.50 for prompts above 100K tokens, five times its short-prompt price. Luna vs Haiku 5.5 in detail.
- Not caching. Cache reads are $0.10 per million tokens for both GPT-6.1 Sol and Sonnet 5.5, and $0.01 for both GPT-6 Luna and Haiku 5.5 on short prompts. The ladder goes all the way down.
- Not the clouds either. In our tracked data, AWS Bedrock, Azure and Google Vertex AI list Claude Sonnet 5.5 at the same $2 / $10, and Bedrock and Azure list GPT-6.1 Sol at $2 / $10. Unlike open-weight models, where hosts differ by up to 7.8x, there is no cheaper reseller for these.
Matching a rival's list price is normal in any market, and nothing here says how much large customers pay after negotiation. For everyone on list price it means one thing: the lab is not your cost lever.
What I would do
- Pick the step first, the lab second. Most features do not need the $10 / $50 step. Run your evals on a $2 / $10 model before you pay five times as much.
- On the $2 / $10 step: test Sonnet 5.5, GPT-6.1 Sol and Gemini 4 Argon in November. From December 1, Argon costs twice as much as the other two (why that matters).
- Before paying for Claude Opus 5.5: compare it with Sonnet 5.5 on your own tasks. On GPQA the cheaper model wins.
- If cost matters most: Gemini 3.8 Flash costs $15 for the month above ($30 from January), GPT-6 Luna and Haiku 5.5 cost $2.
Price your own volume on any model page, or compare Claude Opus 5.5 vs Gemini 4 Argon and GPT-6.1 Sol vs Gemini 4 Argon.
The prices behind this take
Checked by hand on November 8, 2026: OpenAI API pricing (GPT-6 Astra, GPT-6.1 Sol, GPT-6.1 Terra and GPT-6 Luna, short and long context, cached input), Claude pricing (Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, including Haiku's 100K threshold and cache reads) and Gemini API pricing (the dated rates for Gemini 4 Argon and Gemini 3.8 Flash). Accuracy scores: Epoch AI benchmarks, GPQA Diamond, at the reasoning setting shown. Cloud prices: our daily tracked data.
The honest summary: at list price, OpenAI and Anthropic sell the same tiers at the same prices, and Google is moving onto the same steps. Choose the step on your own evals, then choose the lab on quality, limits and terms.
Today's API prices
Updated daily from our price tracking; the numbers above are frozen at publication.
| Model | Cheapest provider | Input / 1M | Output / 1M | Providers |
|---|---|---|---|---|
| GPT-6.1 Sol | OpenAI | $2.00 | $10.00 | 4 |
| Claude Sonnet 5.5 | Databricks | $2.00 | $10.00 | 5 |
| Claude Opus 5.5 | Anthropic | $4.00 | $20.00 | 6 |
| Gemini 4 Argon | Google Gemini API | $2.00 | $10.00 | 4 |
Opinion, with receipts
More No-BS Takes
This is an opinion piece. The numbers are sourced; the conclusions are the author's. No provider paid for or reviewed it.