· CostMyTokens · Analysis · 2 min read
OpenAI vs Anthropic vs Google: LLM API pricing compared (October 2026)
Three providers, three pricing philosophies. Where each one is cheapest, where the fine print lives, and why the answer flips with your workload.
As of late October 2026, the three major LLM APIs have settled into recognizably different pricing philosophies. OpenAI sells a ladder. Anthropic sells discipline. Google sells promotions. Which one is cheapest depends less on the sticker price than on the shape of your workload.
The headline numbers
Official list prices per 1M tokens (input / output), verified October 24, 2026:
| Tier | OpenAI | Anthropic | |
|---|---|---|---|
| Frontier | GPT-6 Astra — $10 / $50 | Claude Opus 5.5 — $4 / $20 | Gemini 4 Argon — $2 / $10, then $4 / $20 from Dec 1 |
| Mainstream | GPT-6.1 Sol — $2 / $10 | Claude Sonnet 5.5 — $2 / $10 | Gemini 3.1 Pro — $2 / $12 |
| Volume | GPT-6 Luna — $0.10 / $0.50 | Claude Haiku 5.5 — $0.10 / $0.50 | Gemini 3.8 Flash — $0.75 / $3.75 |
At the mainstream tier the three providers have effectively converged: Sol, Sonnet and (until December) Argon all sit at $2/$10. When headline prices tie, the fine print decides.
Where each provider wins
OpenAI wins on caching economics at the top end and on tokenizer transparency. Sol’s cached input at $0.10 is 95% off, and its prices are the easiest to verify because the pricing page is public and detailed. It loses on complexity: speed tiers multiply rates, and the long-context segment above ~272K input tokens roughly doubles input prices.
Anthropic wins on predictability. Three models, batch at half price across the board, cached reads at a tenth of input. Its one sharp edge is Haiku’s 100K-token cliff, where the rate quintuples for the whole prompt — cheap until the day it is not.
Google wins the sticker-price war while promotions last — Argon at $2/$10 is a frontier model at a mainstream price, and its 95% cache discount matches the best anywhere. The catch is the calendar: Argon doubles December 1 and the Flash line doubles January 1, and output bills include thinking tokens you never see.
The verdict is workload-shaped
- Chat at moderate volume: a three-way tie at $2/$10; choose on quality and ecosystem.
- Batch pipelines: Anthropic’s across-the-board 50% batch discount is the simplest big lever.
- Cache-heavy agents: compare cached rates first — Sol and Argon at $0.10 cached beat models with lower headline prices but no published cache rate.
- Very long prompts: model the tier thresholds (Haiku’s cliff, Sol’s segment, Gemini 3.1 Pro’s 200K break) before the headline rate.
Run your own mix through the calculator — every comparison on this site, including this article’s tables, is generated from the same dated, sourced dataset, and the head-to-head pages recompute their verdicts from it.