1300 166 138 Tell us what you need
News / OpenAI

GPT 5.6 Sol now costs less to run than Claude Opus 5

By Monika Tantau • Published 3 months ago
GPT 5.6 Sol now costs less to run than Claude Opus 5

Key takeaways

• OpenAI cut GPT 5.6 Sol to $5.60 per million input tokens and $28 output on 21 August 2026
• Output fell 33 per cent and input 20 per cent, converted at the Reserve Bank rate for that day
• Sol now undercuts Claude Opus 5 on both input and output for the first time
• The pricing is promotional and runs at least until 21 November 2026

Most Read

OpenAI dropped the price of its flagship model on Friday. GPT 5.6 Sol now costs $5.60 per million input tokens and $28 per million output tokens, down from $7 and $42. That is a fifth off the input rate and a third off output.

The cut is promotional and runs at least until 21 November 2026. It also does something the rate card has not done before: it puts OpenAI’s flagship underneath Anthropic’s Claude Opus 5 on both sides of the meter.

What the new rate card says

Input fell from $7 to $5.60 per million tokens and cached input from 70 cents to 56 cents, according to OpenAI’s announcement. Output took the deeper cut, from $42 to $28. Every figure here is converted from the published United States rate at the Reserve Bank of Australia rate for 21 August 2026, when one Australian dollar bought 71.45 American cents.

The change applies to the API and is rolling out across eligible plans for Codex credits and ChatGPT Work, so teams on those plans get more work from the same credit balance. Pro, Plus and Business subscription usage is unchanged. OpenAI framed the move as efficiency gains passed through while it keeps pushing capability.

Sol now sits under Opus 5

Claude Opus 5 lists at $7 per million input tokens and $35 output at the same day’s rate. Sol at $5.60 and $28 undercuts it by 20 per cent on both sides. Until Friday the two flagships shared an input price and Opus 5 was the cheaper of the pair on output.

That reversal matters more than the headline percentage. Teams who picked Opus 5 on cost rather than capability made a call that no longer holds at list price, and the gap between Opus 5 against Sol now runs the other way. Claude Fable 5 is untouched at $14 and $70, still the premium option by a wide margin.

The clock runs out in November

This is a promotion, not a rate cut. OpenAI has committed to the new prices at least through 21 November 2026 and has said nothing about what follows. Anyone rebuilding a cost model on $28 output is budgeting against a number that has an expiry date on it.

If the price reverts, output goes back to $42, a 50 per cent increase on what a team would then be paying. Three months is long enough to migrate a production workload onto Sol and short enough that the migration finishes just as the rate changes.

Why Australian businesses should care

OpenAI bills in United States dollars, so an Australian invoice has two moving parts rather than one. The rate card is promotional and the currency floats underneath it. At the Reserve Bank rate for 21 August, a million input tokens plus a million output costs $34, down from $49. A five cent move in the currency shifts that by roughly two dollars either way.

For most Australian operators the useful question is not which flagship is cheapest this week but whether the workload belongs on a flagship at all. Terra and Luna already absorb most routine business volume, and nothing on Friday changed the case for keeping classification or extraction work off the top tier. Sorting that out is ordinary AI consulting work, and it moves the bill further than a 20 per cent promotion does.

What to watch before November

What to watch before November

Terra and Luna are unchanged. Both were cut on 30 July 2026 and neither moved on Friday, which leaves the volume tiers where most Australian workloads already sit. Our comparison of Sol, Terra and Luna sets out where each tier earns its price.

The framing differs depending on who describes it. OpenAI put the cut down to efficiency gains, while the competitive read, as Reuters reported on Friday, is pressure from Anthropic and from Chinese models. Both can be true, and the expiry date is the tell: a permanent efficiency gain does not usually need one.

The number worth watching is not the price but what OpenAI does with it on 21 November. If the promotion becomes the list price, the flagship tier has genuinely repriced and every comparison written before Friday needs redoing. If it lapses, this was a three month campaign against a competitor, and the teams that migrated will find out what that cost them.

Scroll to Top