Claude Opus 5.5 vs GPT-6 Sol vs Gemini: pricing and what each is for
September 25, 2026 · JICA LABS
Three big launches landed in under a month: Gemini 3.8 Flash (September 2), Claude Opus 5.5 and GPT-6 Sol and Luna (both September 22). If you plan to use one in an app, a bot or your work, the practical question is what each costs and what it is good for. Here are their official API prices as of September 25, 2026, a worked example with real numbers and a calculator for your own case.
First: how they charge (tokens)
AI APIs do not charge per message but per token, small chunks of text. As a rule of thumb, OpenAI estimates one English token is about 4 characters, or three quarters of a word. They charge separately for:
- Input: everything you send the model (your question, instructions, conversation history, documents).
- Output: what the model writes back. At all three providers output costs 5 to 6 times more than input.
Price table (USD per million tokens)
| Model | Company | Input | Output |
|---|---|---|---|
| Most expensive | |||
| Claude Fable 5.1 | Anthropic | $10 | $50 |
| GPT-6 Astra | OpenAI | $10 | $50 |
| Mid-range | |||
| Claude Opus 5.5 | Anthropic | $4 | $20 |
| Gemini 3.1 Pro (preview) | $2 | $12 | |
| Claude Sonnet 5 | Anthropic | $2 | $10 |
| GPT-6 Sol | OpenAI | $2 | $10 |
| Budget | |||
| Claude Haiku 4.5 | Anthropic | $1 | $5 |
| Gemini 3.8 Flash * | $0.75 | $3.75 | |
| Gemini 3.1 Flash-Lite | $0.25 | $1.50 | |
| GPT-6 Luna | OpenAI | $0.10 | $0.50 |
* Gemini 3.8 Flash: launch price until December 31, 2026; from January 1, 2027 it becomes $1.50 / $7.50. These are standard rates for normal-length prompts. For very long prompts OpenAI and Google charge more: for example, GPT-6 Sol rises to $4 / $15 and Gemini 3.1 Pro to $4 / $18. Anthropic charges much less when you reuse the same context (caching): cache reads on Opus 5.5 cost $0.20 per million instead of $4.
What it costs in practice
Example: an assistant answering 1,000 requests a month, each with about 2,000 input tokens (question + instructions + some history) and 500 output tokens. That adds up to 2 million input and half a million output tokens per month:
| Model | Cost per month |
|---|---|
| GPT-6 Luna | $0.45 |
| Gemini 3.1 Flash-Lite | $1.25 |
| Gemini 3.8 Flash * | $3.38 |
| Claude Haiku 4.5 | $4.50 |
| Claude Sonnet 5 | $9.00 |
| GPT-6 Sol | $9.00 |
| Gemini 3.1 Pro (preview) | $10.00 |
| Claude Opus 5.5 | $18.00 |
| Claude Fable 5.1 | $45.00 |
| GPT-6 Astra | $45.00 |
The gap is huge: what costs $0.45 with GPT-6 Luna costs $18.00 with Claude Opus 5.5 and $45.00 with GPT-6 Astra. But price per token is not everything: a cheap model that gets things wrong and needs retries can end up costing more per task done right.
Calculator: your case, in your currency
Pick an example or type your own numbers. Exchange rates load in your browser; prices are the ones in the table.
Loading the calculator…
What each one is for, according to each company
- Claude Opus 5.5 (Anthropic): according to Anthropic, it performs at the level of Claude Fable 5.1, its largest model, on most work, and beats it at agentic coding, knowledge work, computer use and chart reading. It is also over 30% faster than Opus 5. More in our Opus 5.5 note.
- GPT-6 Sol (OpenAI): the mid-tier model, with more reasoning, for day-to-day tasks. OpenAI says it makes about half as many mistakes as GPT-5.6 Sol.
- GPT-6 Luna (OpenAI): the cheapest and fastest, built for high-volume responses. GPT-6 Astra is the series' large model. More in our GPT-6 Sol and Luna note.
- Gemini 3.8 Flash (Google): Google highlights gains in software engineering, agentic tasks and multi-step reasoning, and says it outperforms most larger frontier models on long-horizon coding at a much lower cost.
- Gemini 3.1 Pro (Google): Google's newest Pro model, in preview, aimed at understanding images, audio and video, agentic tasks and coding.
Note: these are each company's claims about its own tests. Each one picks what to compare against, so take them with caution.
What if I just want to chat?
Using these models in their apps, without coding, is priced differently:
- Claude: has a free plan. Pro costs $20 a month ($17 a month billed yearly). Opus 5.5 is on the Pro, Max, Team and Enterprise plans.
- ChatGPT: Sol and Luna are on most paid plans; Luna also reaches the Free and Go plans through the desktop app.
- Gemini: the free plan includes Gemini 3.6 Flash and limited access to 3.1 Pro. Gemini 3.8 Flash is in the app for Pro and Ultra subscribers. Google's prices vary by country.
Which one fits what you do
Our practical take, at today's prices:
- Testing an idea or running high volume (a support bot, sorting messages): start with GPT-6 Luna, Gemini 3.1 Flash-Lite or Gemini 3.8 Flash. Move up only if the answers fall short.
- Balanced general use: GPT-6 Sol and Claude Sonnet 5 cost the same ($2 / $10); try both on your own cases.
- Agentic coding and long tasks: exactly where Anthropic positions Opus 5.5 and Google positions Gemini 3.8 Flash. Opus costs more than five times as much per token; check whether the quality gap pays off in your project.
- Always: test with 20 or 30 real cases of your own and measure cost per task done right, not price per token.
Want to call these APIs from your own code? Start with our free Python lessons (in Spanish): the JSON lesson is the basis for reading their responses.
Sources
- Anthropic — Precios de Claude / Claude pricing
- OpenAI — API pricing
- Google — Gemini Developer API pricing
- Google — Introducing Gemini 3.8 Flash and 3.8 Flash Cyber
- Google — Planes de Google AI / Google AI plans
- Anthropic — Claude Opus 5.5
- TechCrunch — OpenAI launches GPT-6 Sol and Luna
- OpenAI Help — What are tokens and how to count them?