OpenAI, Anthropic, and Google Just Converged on the Same Price: $2 per Million Tokens
Three labs that don't coordinate pricing landed on the identical rate for the model tier most people and most businesses actually use day to day. That's worth noticing on its own, separate from whichever model is marginally better this month.
The timing makes it easier to see. GPT-6.1 Sol shipped September 29, Gemini 4 Argon's pricing was announced the next day, and Claude Sonnet 5.5 has been running at $2/$10 since its own release. Within 48 hours, the industry's three biggest labs were all quoting the same number for their mid-tier model.
What a million tokens actually buys
A token is roughly ¾ of an English word, so 1,000 tokens works out to about 750 words, per OpenAI's own tokenizer documentation. A million input tokens is therefore around 750,000 words of text fed into the model, close to the length of the entire Lord of the Rings trilogy, for $2.
Where the money actually lands
| Model | Input / MTok | Output / MTok | Cached input / MTok |
|---|---|---|---|
| GPT-6.1 Sol | $2 | $10 | $0.10 |
| GPT-6 Astra | $10 | $50 | $1 |
| Claude Sonnet 5.5 | $2 | $10 | $0.20 |
| Claude Opus 5.5 | $4 | $20 | $0.20 |
| Gemini 4 Argon (intro) | $2 | $10 | $0.10 |
| Gemini 4 Argon (standard, date unset) | $4 | $20 | $0.20 |
Sources: OpenAI, Anthropic, and Google. Argon isn't actually purchasable yet outside Google's Fairwind Program, so its row is a published price without a working checkout.
Sol nearly matches its own flagship on real coding work
The reason the $2/$10 tier is credible now, rather than just the cheap option, is that it's catching up on actual performance. On DeepSWE v1.1, a benchmark built from real software bug fixes rather than puzzle-style tests, GPT-6.1 Sol scored 75.2% against GPT-6 Astra's 74.8%, per MarkTechPost's coverage of the release. Sol also ships with a 1,050,000-token input context window, larger than Astra's.
OpenAI hasn't published exactly how Sol closes that gap at a fifth of the price. What's verifiable is the outcome: a cheaper model beating, not just approaching, its own company's most expensive one on a benchmark built from real bug fixes.
Picking a tier
Occasional chat, writing, research
Light, irregular use where token volume stays low regardless of price.
- ✦Any current model's free or entry-level tier covers this
- ✦The per-token price differences here barely register at low volume
- ✦Don't pay flagship rates for this use case
Daily coding agents, heavy professional use
Where token volume is high enough that price per million actually shows up on a bill.
- ✦The $2/$10 tier (Sol, Sonnet 5.5) is the rational default now
- ✦Near-flagship coding performance at a fifth to a half the price
- ✦Gemini 4 Argon's matching intro rate isn't usable yet outside Fairwind
Maximum capability, cost secondary
Research, high-stakes drafting, or tasks where being wrong costs more than the token bill.
- ✦GPT-6 Astra and Claude Opus 5.5 still lead on the hardest tasks
- ✦Pay flagship prices only when the task actually needs flagship accuracy
- ✦Most daily work doesn't clear that bar
Is GPT-6.1 Sol the same price as Claude Sonnet 5.5?
Yes. Both charge $2 per million input tokens and $10 per million output tokens on their standard API rates, as of October 2026.
Can I actually buy Gemini 4 Argon at its $2/$10 introductory price?
Not yet for most people. Google is rolling Argon out first to vetted organizations in its Fairwind Program; paid API customers and Google AI Ultra subscribers are next in line, with no date confirmed.
Is GPT-6.1 Sol actually as good as GPT-6 Astra?
On DeepSWE v1.1, a benchmark built from real software bug fixes, Sol scored 75.2% against Astra's 74.8%, essentially matching it on that measure while costing a fifth as much per token.
What does a 'cached input token' mean for my bill?
When a model reuses a portion of a prompt it has already processed (a system prompt or long document sent repeatedly, for example), providers charge a steep discount on that repeated portion rather than the full input rate.
Is Claude Opus 5.5 worth paying double or more over Sonnet 5.5?
Only for tasks that specifically need the higher tier's extra accuracy. For most day-to-day coding and writing work, the $2/$10 tier now performs close enough that the price difference is hard to justify.
Try the tools ✦
Free browser tools that never upload your files.