Home → Guides → GPT-6.1 Sol vs Claude Opus 5.5: same answers, very different bills

GPT-6.1 Sol vs Claude Opus 5.5: same answers, very different bills

Two of the most-asked-about models this month, put through identical prompts with answers a script can check. The interesting result is not who got it right — both did — but how much each one wrote to get there.

List prices

Per million tokens. gpt-6.1-sol is billed at the same rate as gpt-6-sol; Opus 5.5 is from Anthropic's pricing page.

Model	Input	Cache read	Output
gpt-6.1-sol	$2.00	$0.20	$10.00
claude-opus-5-5	$4.00	$0.20	$20.00

On paper Opus 5.5 costs exactly twice as much. What the same models cost here, taken live from the current rates:

Model	Input / 1M	Output / 1M
gpt-6.1-sol	$0.400	$2.00
claude-opus-5-5	$0.900	$4.50

The test

Three prompts, each with an answer a script can verify, run twice per model through this gateway on 10 October 2026:

• Logic — a five-person seating puzzle with exactly one solution, answer required in a fixed format.
• Extraction — pull three orders out of a messy paragraph into JSON, with quantities written as words in one case.
• Code — write merge_intervals() plus four asserts; the code is executed, and passes only if every assert holds.

Results

Task	Correct	Output tokens 6.1 Sol	Output tokens Opus 5.5	Output cost at list price 6.1 Sol	Output cost at list price Opus 5.5
Logic	4 / 4	73–154	441–794	$0.0007–0.0015	$0.0088–0.0159
Extraction	4 / 4	67–85	116	$0.0007–0.0009	$0.0023
Code	4 / 4	202–203	372–456	$0.0020	$0.0074–0.0091

Output cost is output tokens × each model’s list output rate, so it isolates how much each model chose to write. The price you pay here is that figure times the rate in the table above.

Why the gap is wider than the price list says

The rate card says 2×. Measured on output, the gap was about 3× on extraction, 4× on the code task and up to 20× on the logic puzzle. The difference is length: on the puzzle Opus 5.5 walked through every case before answering, while 6.1 Sol stated the seat and stopped. Both were right. When you pay per output token, a model that explains itself costs more even when the explanation adds nothing your code will read.

That cuts both ways. If a person reads the answer, or you need the reasoning to audit it, Opus 5.5's longer working is worth paying for. If the output goes straight into a parser, you are paying for prose nobody sees.

How to route between them

• Machine-read output — extraction, classification, tool calls, short code: gpt-6.1-sol. Same correctness on these tasks at a small fraction of the cost.
• Human-read output — reviews, explanations, long-form writing, anything where the working matters: claude-opus-5-5.
• Long agent sessions: both charge $0.20 per million for cache reads, so a stable, cached prefix costs the same on either. The difference comes down to how much each writes per turn.

Three tasks is a small sample, and it says nothing about hard reasoning or long documents — run your own prompts through both. Per-model rates in 18 languages are on the gpt-6.1-sol and claude-opus-5-5 pricing pages.

Start using it

No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.

Read the 30-second quickstart

Prices quoted on this page are regenerated automatically from live billing data. Third-party terms are quoted from that party's own published documentation.