GPT-6.1 Sol Pro: The Listing Arrived in Hours, the Setting Arrived Months Ago

GPT-6.1 Sol Pro is not a model you can buy from Open​AI, and the honest version of this week's story is that it never was. The name resolves to GPT-6.1 Sol — released on September 29, 2026 — with reasoning.mode set to pro in the Responses API, which Open​AI documents as a mode of an existing model rather than a product with its own identifier. What gives the name its news value is the second half: within hours of GPT-6.1 Sol going live, pro listings for it appeared in third-party catalogues next to older siblings — GPT-6 Sol Pro, GPT-6 Luna Pro, GPT-5.6 Sol Pro — and all of them quote the base model's own rate card. Read that as a launch and you will plan a migration around a product that does not exist. Read it as a mode and you get something more useful: the cheapest documented way to run a pro configuration Open​AI has ever shipped, and a billing shape that is worth understanding before you switch anything to it.

OpenAI's own deprecation table already told you pro is a mode

The cleanest evidence is not a blog post or a forum thread; it is the substitute column in Open​AI's deprecations page, where the vendor writes its own migration advice. When the older Pro meters shut down, the recommended replacement is not a new Pro identifier. It is a model plus a mode.

• Shutting down October 23, 2026 — o1-pro-2025-03-19 and o1-pro are replaced by gpt-5.6-sol with reasoning.mode: pro
• Shutting down December 11, 2026 — gpt-5-pro-2025-10-06 is replaced by gpt-5.6-sol with reasoning.mode: pro
• Shutting down December 11, 2026 — o3-pro-2025-06-10 is replaced by gpt-5.6-sol with reasoning.mode: pro

That is the vendor telling developers, in a table, that the thing they used to buy as a premium model now arrives as a parameter. It also explains the shape of the price list. The Pro identifiers that remain are a closed set from a previous era, priced on their own meters and getting no successors: gpt-5-pro at $15.00 input and $120.00 output per million tokens, gpt-5.2-pro at $21.00 and $168.00, gpt-5.4-pro and gpt-5.5-pro at $30.00 and $180.00 on short context, o3-pro at $20.00 and $80.00, and o1-pro at $150.00 and $600.00. Not one model in the GPT-5.6 or GPT-6 families appears there with a -pro suffix, and a search of the vendor's public API schema returns four GPT-6-era identifiers — gpt-6-astra, gpt-6-sol, gpt-6-luna and gpt-6.1-sol — with no pro variant among them.

What actually changed on September 29

Two things moved this week, and only one of them is about the model. The first is GPT-6.1 Sol itself: a single-snapshot tier at gpt-6.1-sol, 1,050,000 tokens of context, 128,000 maximum output tokens, an April 30, 2026 knowledge cutoff, and the same $2.00 / $10.00 headline rate as the generation it replaces. The second is smaller and easier to miss — the reasoning guide's worked example for pro mode now calls gpt-6.1-sol by name:

• The example request — a curl call to /v1/responses with "model": "gpt-6.1-sol" and "reasoning": {"mode": "pro", "effort": "medium"}
• What that settles — pro mode is documented against the 6.1 identifier by the vendor, in the vendor's own guide, not inferred from a third-party listing
• What it does not settle — whether the mode is validated on other GPT-6-era identifiers. The guide scopes pro mode to "GPT-5.6 and GPT-6 models" as a family, and Open​AI has published no per-model eligibility list for it

There is no separate GPT-6.1 Sol Pro page, no second pricing row, and no announcement. The mode is documented; the product is not.

The billing shape, in the only unit OpenAI publishes

Here is the sentence that decides what a pro configuration costs, and it is worth reading twice because it is the opposite of how the old Pro meters worked: pro mode "aggregates the model work performed to produce the final answer and bills those tokens at the selected model's standard token rates." There is no pro premium on the rate card. The surcharge arrives as volume — more tokens, at the ordinary per-token price — and Open​AI does not publish the multiplier, only the direction. The vendor's own guidance is that pro mode is for cases where a marginal quality improvement materially changes the outcome, and that standard mode is preferred for routine, latency-sensitive or high-volume work.

That structure is why the cached-input rate is the number to look at rather than the headline price, and it is the one place the 6.1 tier genuinely changes the arithmetic:

• Input and output — $2.00 and $10.00 per million tokens, identical on gpt-6.1-sol and gpt-6-sol, so the pro mode's extra tokens cost the same per unit on either
• Cached input — $0.10 per million tokens on 6.1 Sol against $0.20 on 6 Sol, with cache writes at $2.50 on both. Pro mode performs its extra work inside a request that still carries whatever prefix you sent, so an agent loop that replays a stable prefix thousands of times pays half as much for that prefix while running pro
• Long prompts — past 272,000 input tokens the whole request reprices at 2× input and cache rates and 1.5× output, on both tiers, and that multiplier lands on the pro mode's extra volume too
• Two documented traps — setting reasoning.effort to none in pro mode returns an error rather than being ignored, and the extra reasoning tokens are reported under output_tokens_details.reasoning_tokens without being returned in the response body, so a client that estimates cost from the response text will under-count pro usage every time

Put those together and the practical claim is narrow but real: for the same task at the same effort, a pro configuration on the 6.1 tier is strictly cheaper than the same configuration on the 6.0 tier, and the saving is proportional to how much of your traffic is reused prefix. Everything else about the comparison is a measurement you have to run, because the multiplier is unpublished.

How to resolve any pro listing in two minutes

Three checks settle the question for any generation, and they take less time than reading the listing's marketing copy.

• Look for the identifier on the vendor's own model page. If the vendor documents one snapshot and no pro variant — as the GPT-6.1 Sol page does, listing gpt-6.1-sol and nothing else — the pro entry is a configuration, not a product
• Compare the quoted price to the base model's own rate card. An identical line means the listing is the same model in a different mode and there is no second meter to pay. A line several times the base rate means a genuine premium meter, which is what the older Pro identifiers were
• Check the date on the listing against the model's release date. A pro entry that appears within hours of the underling model's launch, quoting that model's standard rates, was generated from the catalogue row rather than announced by the vendor. The pro listing for GPT-6.1 Sol and the model itself carry the same release date — September 29, 2026 — and that coincidence is the tell

One caveat about independent evidence: there is none yet for this tier. The neutral board this blog uses as its third-party reference has no GPT-6.1 Sol page at all; its model URL for the slug returns a 404 and the string does not appear in the live leaderboard, so any comparative score you see for the 6.1 tier today is being quoted from the vendor against the vendor. That is not a reason to distrust the release. It is a reason to run your own task set before committing a production path to the new tier, pro mode or otherwise.

What you can actually call today

On our side, plainly: neither GPT-6.1 Sol nor any pro listing of it is on OrcaRouter's routes today. The public catalogue returns "model not found" for both openai/gpt-6.1-sol and openai/gpt-6.1-sol-pro, and our Open​AI slate carries GPT-6 Astra, GPT-6 Sol and GPT-6 Luna. What is callable is the model the pro mode is documented against in the current generation — openai/gpt-6-sol at Open​AI's own list price of $2.00 input and $10.00 output per million tokens with $0.20 cached, passed through with nothing added, including the 272K repricing rule. That pass-through is the reason a mode-versus-product distinction matters on our side too: when Open​AI changes a rate, the change is live on the route the same day, with no second contract to renegotiate and no per-model re-onboarding, because there is only one price list and we do not mark it up. When the 6.1 tier lands it will appear in the catalogue the same way, under one API key alongside the 200-plus models already routable, and a pro configuration will be a parameter on that request rather than a second route.

So the decision for this week is smaller than the listing count suggests. If what you want is a pro configuration on Open​AI's newest Sol weights, there is nothing to order and nothing to wait for beyond the tier becoming routable where you already call your models — the mode is documented, the rates are the base model's, and the only thing a "GPT-6.1 Sol Pro" entry adds is a name. If what you want is a pro configuration today, the documented one is reasoning.mode: pro on a GPT-6-era model, and the arithmetic is friendlier on the 6.1 tier and friendlier still on a route that passes the vendor's cache rate through unmodified. The name in the catalogue is the least informative part of the whole question.