Opus 5.5 vs GPT-6 Sol: what we can compare before either ships

Both models arrived on the same Tuesday, hours apart. Anthropic released Claude Opus 5.5 and OpenAI released GPT-6 Sol on 22 September 2026, and each sits one tier below its company's flagship, the level most paying subscribers use every day.

Status on 22 September 2026 at 21:30 UTC: both models are released. Opus 5.5 is on the Claude API as claude-opus-5-5 and GPT-6 Sol on the OpenAI API as gpt-6-sol. The first independent scores, from Artificial Analysis, were published on launch day.

The short version: Opus 5.5 is the stronger model on independent tests, 58 against 48 on Artificial Analysis' Intelligence Index at max effort. GPT-6 Sol costs half as much per token, writes far fewer tokens per task and generates text faster. Which one is better value depends on how hard your work is. The launch details are in our pages on Claude Opus 5.5 and GPT-6 Sol.

The specs side by side

Prices and specs from Anthropic's and OpenAI's own pages, checked 22 September 2026.

Independent scores

From the Artificial Analysis comparison, checked 22 September 2026. Opus 5.5 was tested with Anthropic's default fallback for requests its safeguards block.

Opus 5.5 leads on all ten tests in the index, by the widest margins on agentic knowledge work and coding. On long-context reading, document reasoning and physics the two are within a point. The cost picture runs the other way: Sol charges half as much per token and wrote about a quarter as many output tokens per task, so the same job costs far less on Sol. Artificial Analysis puts Opus 5.5 at $5.98 per index task at max effort and has not yet published a figure for Sol.

What each company claims

Neither launch chart puts these two models head to head. Anthropic compared Opus 5.5 with GPT-6 Astra and GPT-5.6 Sol, and OpenAI compared GPT-6 Sol with Claude Opus 5 and Claude Fable 5.1.

• Anthropic says Opus 5.5 at its default effort beats GPT-6 Astra on the FrontierCode coding benchmark "at roughly 20% of the cost per task", and on GDPval-AA beats Astra at max effort "for about a fifth of the cost per task".

• OpenAI says GPT-6 Sol at xhigh effort beats Claude Opus 5 at max effort on AutomationBench "at just 9% of Opus 5's cost per task", and comes within 1.1 points of Claude Fable 5's best on DeepSWE v1.1.

Both pitches are about cost as much as capability. Anthropic's is that Opus 5.5 does flagship work at a lower price than a flagship. OpenAI's is that Sol does most of it at a fraction of Claude's price.

Same tier, different prices

The tiers above them are priced alike. Claude Fable 5.1 and GPT-6 Astra both cost $10 per million input tokens and $50 per million output tokens, and both are hard to reach on entry-level plans. Anthropic's support page says that on Claude Pro, "Fable 5 and Fable 5.1 aren't included in your plan's usage limits", and OpenAI's help centre says ChatGPT Plus includes Astra only in ChatGPT Work and Codex.

One tier down, the two companies went different ways. Anthropic cut Opus by 20% per token. OpenAI halved Sol's price. At $2 and $10, GPT-6 Sol now costs exactly what Claude Sonnet 5 does, Anthropic's tier below Opus.

How far each one moved

Anthropic spent its new generation mostly on capability, and OpenAI mostly on price. OpenAI also says Sol makes about half as many factual mistakes as GPT-5.6 Sol on its internal factuality evaluation, an improvement the index does not fully capture.

Which should you use?

Reach for Opus 5.5 for your hardest work: long research reports, complex analysis, careful writing, multi-step problems where a wrong turn costs you time. It scored higher on every independent test above, and Anthropic's testers singled out clearer writing.

Reach for GPT-6 Sol when you need speed and volume: everyday questions, drafts, quick analysis, long back-and-forth sessions. It is fast and frugal with tokens, and for everyday tasks that may matter more to you than the quality gap.

The best test is still your own work:

1. Pick three real tasks from the past week: a document to summarise, a piece of writing to fix, a question to research.

2. Give each model the same prompt, in a fresh conversation, with the same files attached.

3. Read the answers before checking which model wrote which, if you can manage it.

4. Time the wait as well. Speed is a large part of Sol's pitch.

5. Check the limits on your plan. A model you can reach all day is easier to rely on than one you can use a few times.

We are adding both models to MultiChats on our paid plans. Once they are in your model picker, you can ask Opus 5.5 a question, branch the conversation into a new thread and switch that thread to GPT-6 Sol, so each model answers from the same history.

Frequently asked questions

Is Opus 5.5 better than GPT-6 Sol?

On independent tests, yes. Artificial Analysis scores Claude Opus 5.5 at 58 and GPT-6 Sol at 48 on its Intelligence Index, both at max effort, and Opus 5.5 leads on all ten benchmarks in it, some by a single point.

Which is cheaper, Opus 5.5 or GPT-6 Sol?

GPT-6 Sol. Its API price is $2 per million input tokens and $10 per million output tokens, half of Opus 5.5's $4 and $20, and it used about a quarter as many output tokens per task in Artificial Analysis' tests.

Which is faster?

GPT-6 Sol, on the evidence so far. Artificial Analysis measured about 104 output tokens per second for Sol and has not yet published a speed for Opus 5.5. Anthropic says Opus 5.5 generates output more than 30% faster than Opus 5.

When did Opus 5.5 and GPT-6 Sol come out?

Both on Tuesday 22 September 2026.

Which has the bigger context window?

They are nearly level: 1,000,000 tokens for Opus 5.5 and 1,050,000 for GPT-6 Sol. OpenAI bills prompts above 272,000 input tokens at a higher rate.

Update log

22 September 2026, 21:30 UTC. Both models released. Replaced the leaked claims with published specs and prices, added Artificial Analysis scores and a recommendation.

21 September 2026, 20:22 UTC. Page opened. Neither claude-opus-5-5 nor gpt-6-sol appeared in its company's public model documentation.