Opus 5 beats GPT-5.6 Sol by 4 points for 25% more. At Sol's undiscounted price, it would cost less.
Opus 5 scores 51 to Sol's 47 at $5 / $25 against $4 / $20. Sol's price is promotional, and above 272K tokens it doubles. Where the order flips.
GTM Architect & Growth Operator · Now · 16 September 2026
TL;DR · Key insights
- On Artificial Analysis's Intelligence Index v4.3, Claude Opus 5 scores 51 and GPT-5.6 Sol scores 47. On the Coding Agent Index the gap is 60 against 55.
- At today's list prices Opus 5 costs 25% more: $5 / $25 against $4 / $20. OpenAI describes Sol's price as a 20% and 33% reduction that is promotional through at least 21 November, which implies $5 / $30 undiscounted.
- At that implied price Sol would match Opus 5 on input and cost 20% more on output while scoring four points lower. Above 272K input tokens Sol already costs more than Opus 5 on both sides.
- Sol is the faster model, at 64.7 against 49.4 tokens per second, and cost Artificial Analysis less than half as much to run the full index. On short prompts at today's price, that is a real case for it.
These two sit in the same place in their vendors’ lineups. Anthropic tells you to start with Opus 5 for most work. Sol is the top tier of OpenAI’s GPT-5.6 family, priced and scored above Terra and Luna, with GPT-6 Astra now above all three.
Opus 5 wins the benchmark row. Sol wins the price row. The comparison only gets interesting once you read the small print under Sol’s price.
| Info | Claude Opus 5 | GPT-5.6 Sol |
|---|---|---|
| Input / output per 1M, today | $5 / $25 | $4 / $20 |
| Sol's implied undiscounted price | n/a | $5 / $30 |
| Above 272K input tokens | $5 / $25 | $8 / $30 |
| Cache read per 1M | $0.50 | $0.40 |
| Intelligence Index v4.3 | 51 | 47 |
| Coding Agent Index | 60 | 55 |
| Output speed, tokens/s | 49.4 | 64.7 |
| Cost to run the full index | $7,275 | $3,465 |
| Knowledge cutoff | May 2026 | 16 Feb 2026 |
Prices from Anthropic's pricing page and OpenAI's GPT-5.6 Sol model page. The implied price is arithmetic on OpenAI's stated discount, not a published price. Index, speed and run cost from Artificial Analysis, read on 16 September 2026 against Intelligence Index v4.3; Coding Agent Index from its GPT-6 Astra benchmarking note.
The two highlighted rows are where this comparison is decided, and neither shows up on a benchmark chart.
Four points, both ways
On general intelligence Opus 5 leads 51 to 47. On the Coding Agent Index, which scores models inside a coding harness rather than on isolated questions, it leads 60 to 55.
Both gaps point the same way. They are about the same size, too. One benchmark proves little, but a lead that holds on two different measures is a lead.
Sol has its own real advantages. It is faster, at 64.7 tokens per second against Opus 5’s 49.4. And it cost Artificial Analysis $3,465 to run its full index against $7,275 for Opus 5, less than half, which says something about how economically it gets to its score.
Today’s price, and the one OpenAI’s page implies
Today, Opus 5 costs 25% more. That is $5 / $25 against $4 / $20, on both sides.
OpenAI’s model page adds two sentences. The first calls Sol’s $4 / $20 a 20% reduction in input pricing and a 33% reduction in output pricing. The second says that promotional pricing is available at least through 21 November 2026.
Run the percentages backwards and you get $5 / $30. That is not a published price. OpenAI never says Sol costs that after 21 November; it is only the figure the stated discount is taken from.
At $5 / $30, the comparison inverts on output:
- Input: Sol $5, Opus 5 $5. Level.
- Output: Sol $30, Opus 5 $25. Opus 5 is 17% cheaper.
- Index: Opus 5 still leads by four points.
At today’s promotional price, Sol trades four index points for a 20% discount. At the price its own discount implies, it would cost more than Opus 5 on output and still score lower. Price anything long-lived at both.
Long prompts already flip it
No need to wait for November.
On Sol, a prompt with more than 272K input tokens is billed at 2x input and 1.5x output for the full request. That takes Sol to $8 / $30. Anthropic’s pricing page says models from 4.6 on keep one rate across the whole 1M context window, with a 900k-token request billed at the same per-token rate as a 9k-token one. Opus 5 stays at $5 / $25.
So above 272K input tokens, Opus 5 is both the higher-scoring model and the cheaper one, on input and on output. For whole repositories, long document sets or large retrieval contexts, it is not close.
Where each one wins
| The work | Model | Why |
|---|---|---|
| Complex agentic coding | Claude Opus 5 | Leads the Coding Agent Index 60 to 55, and Anthropic positions it for exactly this. |
| Prompts above 272K input tokens | Claude Opus 5 | Higher score and cheaper on both sides once Sol's multiplier applies. |
| Short prompts, latency matters, priced today | GPT-5.6 Sol | Faster and 20% cheaper at today's promotional price. |
| A product whose margins run past November | Claude Opus 5 | Its price is Anthropic's standard one. Sol's has a date and an implied $5 / $30 behind it. |
| Well-specified, high-volume work | Neither | This is a job for a cheaper tier. Compare Terra or Sonnet 5 before paying flagship prices. |
Opus 5 wins on capability, long context and price certainty. Sol wins on speed and on today's price for short prompts.
If you are already deciding within Anthropic’s lineup, the step above Opus 5 is covered in Opus 5 against Fable 5.1. Within OpenAI’s, the step above Sol is in GPT-6 Astra against Sol. And if the real question is Sol against a cheaper Claude, see Sol against Sonnet 5.
What would change my mind
OpenAI confirming Sol’s price after 21 November. If $4 / $20 becomes the standard price, the output row stops inverting and Sol’s case on short prompts gets much stronger.
A cost-per-completed-task comparison on agentic coding for this pair. Sol’s lower run cost on the full index suggests it reaches its score economically, and on the right workload that could outweigh four points.
A change to OpenAI’s 272K multiplier. It is a pricing policy, and it decides every long-context comparison against Claude.
Questions people asked
Is Claude Opus 5 better than GPT-5.6 Sol?
On published benchmarks, yes: 51 against 47 on the Intelligence Index v4.3 and 60 against 55 on the Coding Agent Index. Sol is faster and cost less than half as much to run the full index.
Which is cheaper, Claude Opus 5 or GPT-5.6 Sol?
Under 272K input tokens at today’s price, Sol: $4 / $20 against $5 / $25. Above 272K, Opus 5, because Sol’s whole request goes to $8 / $30. At Sol’s implied undiscounted $5 / $30, Opus 5 would be cheaper on output for short prompts too.
Is GPT-5.6 Sol’s price promotional?
Yes. OpenAI’s model page says promotional pricing is available at least through 21 November 2026 and calls $4 / $20 a 20% and 33% reduction, which implies $5 / $30. It does not say what Sol costs after that date.
Should I use Claude Opus 5 or GPT-5.6 Sol?
Opus 5 for complex agentic coding, long prompts and anything priced past November. Sol for fast, short-prompt work at today’s price, provided the economics also hold at $5 / $30.
If you are pricing a product on one of these and want a second opinion on the model cost, the contact page is the fastest route to me.
