Claude Opus 5.5 versus GPT-6 Sol: pricing, context windows and billing rules
Anthropic's Claude Opus 5.5 (released 2026-09-22) costs $4 input and $20 output per million tokens with cached reads at $0.20; OpenAI's GPT-6 Sol, released the same day, lists $2 input, $0.20 cached input, $2.50 cache writes and $10 output. Both are callable on QCode under one key. This page compares only published rates and specifications — no benchmark numbers at all.
Updated 2026-09-30
Four recomputable examples
Claude Opus 5.5 · 100,000 input + 20,000 output
Computed from the official table, where each rate is per million tokens: 100,000 × $4 per million = $0.40 input, 20,000 × $20 per million = $0.40 output, total $0.80. Long context not entered, Batch not used.
GPT-6 Sol · the same 100,000 input + 20,000 output
Computed from the official table, each rate per million tokens: 100,000 × $2 per million = $0.20 input, 20,000 × $10 per million = $0.20 output, total $0.40. Against X1, both input and output unit rates differ by a factor of two.
The same absolute cache-read rate, two different conventions
Computed from the official tables: 1,000,000 cached input tokens cost $0.20 at both vendors. The derivation differs — Anthropic states 0.05x on a $4 input, OpenAI states 10% on a $2 input — so the same phrase «cached reads are cheap» lands on different numbers in each table.
Claude Opus 5.5 on Batch: the total comes back to $0.40
Computed from the official Batch table, each rate per million tokens: 100,000 × $2 per million = $0.20 input, 20,000 × $10 per million = $0.20 output, total $0.40 — the same number as X2's GPT-6 Sol standard call. That is one direction of coincidence: before comparing totals across vendors, check whether both requests are in the same pricing tier.
What this page compares — and what it does not
Three published things get compared: unit rates, context and maximum-output numbers, and the rules for cache and long context. Three things do not: which model scores higher (no benchmark numbers are transcribed), whose actual bill is lower (that depends on your cache-hit ratio and prompt shape, not on the table), and which retires first (both vendors publish only a «not sooner than» floor).
Two models released on 2026-09-22
What the vendors published: Anthropic released Claude Opus 5.5 on 2026-09-22, stating «Input and output tokens are $4 and $20 per million, 20% less than Opus 5» and cache reads «$0.20 per million tokens, 60% less than Opus 5». OpenAI released GPT-6 Sol and GPT-6 Luna the same day, with the changelog line «GPT-6 Sol: $2 input, $0.20 cached input, and $10 output.» Both tables state the unit as per million tokens.
Two release timelines and one piece of background
OpenAI's changelog, under 2026-09-22, reads «Released GPT-6 Sol (gpt-6-sol) and GPT-6 Luna (gpt-6-luna)»; on the same day Anthropic released Claude Opus 5.5, and Claude Code's changelog at 2.1.280 states «Added Claude Opus 5.5 (claude-opus-5-5), now the default Opus model».
One piece of background: OpenAI released gpt-6.1-sol on 2026-09-29 at the same $2 input and $10 output as GPT-6 Sol, but with cached input at $0.10 — 5% per OpenAI, against the 10% convention of the older Sol. As of 2026-09-30 QCode does not list gpt-6.1-sol as callable, so this comparison stays Opus 5.5 against Sol.
The date this page was re-checked. On 2026-09-30 neither table had moved: the GPT-6 Sol row survived DevDay unchanged at $2.00 / $0.20 / $2.50 / $10.00, and Anthropic's Opus 5.5 row is still $4 / $5 / $8 / $0.20 / $20. On retirement both vendors only give a floor: Opus 5.5 not sooner than 2027-09-22.
Layers of sourcing: official text / our arithmetic / site measurement
Officially announced rates and specifications
Confirmed side by side (both vendors' tables re-checked 2026-09-30): Claude Opus 5.5 lists a 1M context window, 128K maximum output and a reliable knowledge cutoff written Jun 2026, priced at $4 input, $5 five-minute cache writes, $8 one-hour cache writes, $0.20 cache reads and $20 output. GPT-6 Sol lists 1,050,000 context, 922,000 maximum input, 128,000 maximum output and a knowledge cutoff written Apr 20, 2026, priced at $2.00 input, $0.20 cached input, $2.50 cache writes and $10.00 output. The two cache-read rates are the same $0.20 per million, but derived differently: Anthropic writes «0.05x the base input price» on a $4 input, OpenAI writes «10% of the uncached input token rate» on a $2 input.
Unverified; fields not published by the vendors (left blank)
No rumour and no unverified claim is carried here. Both vendors published a knowledge cutoff — Anthropic's model table writes Jun 2026 for Claude Opus 5.5 and OpenAI's model page writes Apr 20, 2026 for GPT-6 Sol — so that field has moved up into the confirmed section. What stays blank: the extent of each vendor's reasoning-effort documentation for these two models, and any statement about which one is the stronger coder, which would need benchmark numbers this page does not transcribe. One thing not to mix up: the model released on 2026-09-29 is gpt-6.1-sol, a different id from the GPT-6.1 Astra reported cancelled on 2026-09-28; OpenAI never links the two.
The two tables side by side
Claude Opus 5.5 (Anthropic)
One row of the official table carries five numbers: $4 input, $5 five-minute cache writes, $8 one-hour cache writes, $0.20 cache reads, $20 output, all per million tokens. The Batch table is $2 / $10; Fast mode is in research preview at $8 / $40. Cache reads are published as 0.05x the base input price. The model table adds two dates: a reliable knowledge cutoff written Jun 2026, and retirement «Not sooner than September 22, 2027».
GPT-6 Sol (OpenAI)
The short-context tier of the Standard row: $2.00 input, $0.20 cached input, $2.50 cache writes, $10.00 output; above 272K input tokens the whole request moves to $4.00, $0.40, $5.00, $15.00. The model page adds that Batch and Flex are 50% lower, Fast is 2x Standard, regional processing adds 10% where available, and the knowledge cutoff reads Apr 20, 2026.
Switching under one key
On QCode, claude-opus-5-5 and gpt-6-sol are two model ids under the same key — the request body changes the id and nothing else about your endpoint. But both sides have their own conventions: Anthropic's effort and thinking-block structure, and OpenAI's requirement that tool calling go through the Responses API. This page does not compare capability; it lists what has to move when you switch.
The QCode side: both models are in stock
As of 2026-09-30 both claude-opus-5-5 and gpt-6-sol are on QCode's sell list with real 30-day traffic behind them, which is what lets this page say «callable». We do not mark either vendor's rates up or down here; billing follows each vendor's own official table — QCode's own pricing is on the pricing page.
Questions
Is GPT-6 Sol half the price of Opus 5.5?
By the official tables: $2 against $4 for input and $10 against $20 for output, so yes on standard unit rates. The cache and long-context rules differ between the two, however, so an actual bill depends on hit ratio and request length. This page stops at «standard unit rates differ by a factor of two» and does not compute anyone's total.
Whose context window is longer?
Different notations, adjacent numbers: Anthropic's table says 1M tokens, OpenAI's model page says 1,050,000. GPT-6 Sol additionally gives a maximum input of 922,000; the Opus 5.5 page gives a maximum output of 128K — both sit at 128,000 on output.
Who charges more for long context?
The mechanisms differ: OpenAI moves the whole request to 2x input and cache and 1.5x output once input passes 272K (landing on $4.00 / $0.40 / $5.00 / $15.00 in the table), while Anthropic's Opus 5.5 page states cache reads as 0.05x the base input price and gives no equivalent «double above a threshold» sentence in the same place. They have to be described separately; merging them into one line would misrepresent both.
Why are no benchmarks shown?
Because this page keeps to what can be re-checked against the vendors' own tables. Benchmark figures are each vendor's internal testing convention, and none are transcribed here; if you want them, the launch pages carry their score tables — that simply isn't this page's job. Same rule applied to both vendors: verify what is verifiable, leave the rest blank.
Should I switch to gpt-6.1-sol right now?
This page will not make that decision for you; it lays out three facts. First, OpenAI released gpt-6.1-sol on 2026-09-29 at the same standard rates as GPT-6 Sol, with cheaper cached input ($0.10 against $0.20). Second, the official documentation states that gpt-6.1-sol does not support the none or the minimal reasoning effort, while GPT-6 Sol and GPT-6 Luna do support none — and minimal is absent from the published effort list for these two models, which reads none, low, medium, high, xhigh, max. Third, as of 2026-09-30 QCode does not list gpt-6.1-sol as callable.
Will either of these be retired?
Anthropic's deprecations table gives a floor, not a date: the claude-opus-5-5 row reads «Not sooner than September 22, 2027». OpenAI's deprecations page, re-checked on 2026-09-30, contains no entry for gpt-6-sol at all — nor does its October 23 shutdown list.
Sources
Three Anthropic pages: pricing (the table and the 0.05x sentence), the Opus 5.5 launch page (the $4 / $20 and discount sentences) and the deprecations table (the «not sooner than 2027-09-22» row), plus Claude Code changelog 2.1.280. Three OpenAI pages: pricing (the gpt-6-sol and gpt-6.1-sol rows), the GPT-6 Sol model page (specification and billing rules) and the API changelog (the Sep 22 and Sep 29 blocks). Site-side statements use only the 30-day usage record re-checked 2026-09-30; every source line is archived verbatim.
Both vendors under one key
Switch between claude-opus-5-5 and gpt-6-sol with one QCode key (both in stock as of 2026-09-30); the gpt-6.1-sol released on 2026-09-29 is not here yet.
Related pages
Claude Opus 5.5 guide
Opus 5.5's own specification, pricing table and how to call it on QCode — this page only sets the two vendors side by side.
GPT-6 Sol / Luna pricing tracker
The full OpenAI table and its four billing rules — the arithmetic behind X2, X3 and X4 on this page comes from there.
Claude Sonnet 5.5 versus Opus 5.5
The within-vendor comparison: Sonnet 5.5 against Opus 5.5. Cross-vendor comparison is here, sibling comparison is there.
This page transcribes only what the two vendors' own pages state verbatim; every per-call total is our arithmetic from those tables, labelled as a computation rather than a quote or a promise. Vendor statements about discounts and running cost carry their source, and no scores are compared. QCode is independent of Anthropic and OpenAI; the site's usage figures are an observation of 2026-09-30 only and never a service level.