Share This Article
Claude Opus 5.5 is Anthropic’s newest top model, priced at $4 per million input tokens and $20 per million output tokens. GPT-6 Sol is OpenAI’s mid-tier “everyday” model at $2 and $10, half the price. Opus 5.5 is aimed at the hardest coding and knowledge work; Sol is aimed at dependable daily work at low cost. For most marketing teams, the right choice depends less on benchmarks and more on which one does better on your own recurring tasks.
On 22 September 2026, both companies released new models within hours of each other. That makes for easy headlines, but it’s not quite a like-for-like fight: Opus 5.5 is Anthropic’s flagship, while Sol sits below OpenAI’s flagship, GPT-6 Astra. Here’s what each is, what it costs and how to decide.
At a glance
| Claude Opus 5.5 | GPT-6 Sol | |
|---|---|---|
| Made by | Anthropic | OpenAI |
| Released | 22 September 2026 | 22 September 2026 |
| Position in the range | Anthropic’s top generally available model | OpenAI’s mid-tier model (below GPT-6 Astra) |
| API price (per 1M tokens) | $4 input / $20 output | $2 input / $10 output |
| Faster option | Fast mode (research preview) at $8 / $40 | GPT-6 Luna for speed and volume |
| Batch discount | 50% off | Discounts available via caching and batch |
| Where to use it | Claude apps and the Claude API | ChatGPT Work and Codex on paid plans; OpenAI API |
| What its maker emphasises | Coding and knowledge work; “strongest-performing model” Anthropic has tested | Everyday professional work, factuality, low cost |
What’s new in Claude Opus 5.5
Anthropic describes Opus 5.5 as its strongest-performing model to date and says it beats its larger Fable model on many benchmarks. The headline change for buyers is price: Opus 5.5’s output tokens cost $20 per million, down from $25 for Opus 5, and reports put the saving on typical workloads at around 40%. Anthropic also says the model is faster to run.
It also introduced a “fast mode” in research preview on the Claude API, which costs twice as much but responds faster. Anthropic said smaller Sonnet 5.5 and Haiku 5.5 models would follow in the coming weeks.
What’s new in GPT-6 Sol
GPT-6 Sol brings improvements from OpenAI’s flagship GPT-6 Astra into a cheaper model. OpenAI says Sol makes about half as many mistakes as the previous Sol on its internal factuality test, and that its API price is half that of GPT-5.6 Sol. OpenAI also says it adopted Astra’s clearer communication style, with less jargon and slightly shorter answers.
At launch, Sol was available in ChatGPT Work and Codex for paid plans and in the API, but not yet in regular ChatGPT chats. We cover the whole GPT-6 family in our guide to Astra, Sol and Luna.
Is this a fair comparison?
Partly. On price, a closer match for Sol is Anthropic’s Claude Sonnet 5, which also costs $2 input and $10 output. On capability, a closer match for Opus 5.5 is GPT-6 Astra at $10 and $50. So the real question most teams face is:
- Do you want Anthropic’s top model (Opus 5.5) at twice the price of OpenAI’s everyday model (Sol)?
- Or do you compare like-for-like on price (Sonnet 5 vs Sol) or like-for-like on tier (Opus 5.5 vs Astra)?
For marketing work, where most tasks are writing, summarising and analysis rather than hard engineering, the mid-tier models often do the job. Paying for a flagship makes sense when you’ve seen it do meaningfully better on your own work.
What it costs in practice
Take the same example we use in our GPT-6 guide: summarising 1,000 customer reviews, about 500,000 input tokens and 100,000 output tokens in total.
| Model | Approx. cost at list prices |
|---|---|
| GPT-6 Sol | $2.00 |
| Claude Sonnet 5 | $2.00 |
| Claude Opus 5.5 | $4.00 |
| Claude Opus 5.5 (batch) | $2.00 |
| GPT-6 Astra | $10.00 |
These are rough figures. Reasoning tokens, caching and batch processing all change the real bill. For one-off jobs the difference is pennies; at scale, it adds up.
Which to use for which marketing task
We haven’t published head-to-head test results yet, so treat this as a starting point to test, not a verdict.
| Task | Start with | Why |
|---|---|---|
| Long, nuanced copy (brand stories, thought leadership drafts) | Test both | Writing style is personal; both makers claim clearer, more natural output |
| Weekly reports and data summaries | GPT-6 Sol or Claude Sonnet 5 | Routine work where the cheaper tier is usually enough |
| Complex strategy docs from lots of inputs | Claude Opus 5.5 or GPT-6 Astra | Flagship models are built for long, multi-step reasoning |
| Building internal tools, scripts or landing pages | Claude Opus 5.5 | Anthropic positions it strongly for coding |
| High-volume tagging or extraction | GPT-6 Luna or Claude Haiku | Cheapest models are fine for simple, repetitive jobs |
How to run your own 10-task test
Benchmarks measure what model makers choose to measure. Your work is different. This test takes a couple of hours and gives you a real answer.
- Pick 10 tasks you actually do. For example: 5 Google Ads headlines, a product description, a creative brief, a weekly performance summary from a CSV, a blog outline, a customer email reply, a competitor comparison, a social post set, a landing page section, and a keyword grouping job.
- Write one prompt per task and use it word-for-word on both models, with the same files.
- Hide which model wrote which answer (ask a colleague to label them A and B).
- Score each answer 1–5 on accuracy, usefulness and how much editing it needed.
- Record cost and time for each run.
- Total the scores per task. Use the cheaper model wherever it scores within a point of the other.
Our prompt templates for performance marketers make a good starting set, and our editor’s checklist for AI drafts helps you score the writing tasks consistently.
Things to consider beyond the model
- Where your team already works. If your company uses ChatGPT Business, Sol is already in reach. If you use Claude’s apps, Opus 5.5 is.
- Data and privacy settings. Check each provider’s business terms for how your inputs are handled, especially with client data.
- Consistency. Switching models changes tone and output. Pick one default per task and stick to it.
- Human review. Neither model removes the need to check facts, numbers and claims before anything goes live.
Frequently asked questions
Is Claude Opus 5.5 better than GPT-6?
It depends which GPT-6 model and which task. Opus 5.5 is Anthropic’s flagship; GPT-6 Sol is OpenAI’s cheaper mid-tier model; GPT-6 Astra is OpenAI’s flagship. Each company claims leads on different benchmarks. Test on your own work.
How much does Claude Opus 5.5 cost?
In the API, $4 per million input tokens and $20 per million output tokens, with a 50% batch discount. Fast mode costs $8 and $40.
How much does GPT-6 Sol cost?
$2 per million input tokens and $10 per million output tokens in the API.
Which is better for writing marketing copy?
There’s no universal answer. Run the same brief through both and compare how much editing each needs. Many teams end up using one for drafts and another for data work.
Do I need the API to use these models?
No. You can use Opus 5.5 in Claude’s apps and GPT-6 Sol in ChatGPT Work and Codex on paid plans. The API is for building them into your own tools.
The bottom line
Opus 5.5 and GPT-6 Sol aren’t direct rivals so much as two different price points. Use the cheaper tier for routine work, test the flagships on your hardest tasks, and let your own scores decide.
Sources
- Anthropic, Claude pricing
- TechCrunch, Anthropic releases Opus 5.5 with lower prices (22 Sep 2026)
- OpenAI, Introducing GPT-6 Sol and Luna
- Gizmodo, OpenAI releases new GPT-6 Sol and Luna models (Sep 2026)
