OpenAI released GPT 6.1 Sol on 29 September at its DevDay event, one day after Anthropic launched Claude Sonnet 5.5. OpenAI says the new model nearly matches GPT 6 Astra on agentic coding, computer use and professional work at a fifth of Astra’s token price.
The comparison that matters for Australian buyers is closer to home. Both GPT 6.1 Sol and Sonnet 5.5 list at $2.81 per million input tokens and $14.04 per million output tokens, converted at the Reserve Bank rate of 22 September. With the sticker price identical, the choice comes down to how much work each model gets done per dollar.
What GPT 6.1 Sol adds
GPT 6.1 Sol is an upgrade to GPT 6 Sol, launched a week earlier. OpenAI says it matches GPT 6 Astra on DeepSWE, a test of real software engineering tasks, at roughly a fifth of the cost, and scores 2.2 points above Claude Opus 5.5 at medium effort on AutomationBench, a test of multi step business workflows, at about a third of the cost.
Cached input drops to $0.14 per million tokens, half of what GPT 6 Sol charged. OpenAI also reports fewer factual errors on hard prompts: at low effort, the share of answers with an error falls from 11.4% to 7.7%. A faster version called Ultrafast, with up to eight times the output speed in Codex, is due in the coming days.
How it stacks up against Sonnet 5.5
The first independent reading came within hours. On the Artificial Analysis Intelligence Index, GPT 6.1 Sol scores 51.8 at max effort, behind Sonnet 5.5 at 56 and just under GPT 6 Astra at 52.7. At xhigh the two cheaper models almost tie, 51.0 against 51.9, and at high effort GPT 6.1 Sol leads, 50.2 against 46.7.
Cost per task tells a sharper story. Running the index at xhigh costs about $0.55 per task on GPT 6.1 Sol and $3.85 on Sonnet 5.5, because OpenAI’s model uses far fewer tokens to finish. At max effort the gap is $1.02 against $10.67. Sonnet 5.5 still reaches the higher top score, but OpenAI now wins on efficiency at the settings most teams use.

Who can use it, and the rest of DevDay
GPT 6.1 Sol is available now to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, but not yet in the standard chat. Developers can call it through the API as gpt-6.1-sol. OpenAI also launched Dots, always on agents powered by GPT 6 Astra with their own cloud computer, rolling out to Pro, Business Premium and Enterprise plans in eligible markets.
A new ChatGPT Pro 500 plan costs $769.99 a month in Australia and adds Astra Ultrafast, while Pro 200 returns at $300. API usage is still billed in US dollars, so what a local team pays for tokens moves with the exchange rate.
Why Australian businesses should care
Two strong mid tier models at the same list price is good news for buyers. The bill no longer depends on the rate card alone but on how many tokens a model burns to finish a job, and on these first readings that favours GPT 6.1 Sol at everyday settings. Our Sonnet 5.5 vs Opus 5.5 guide shows how much effort settings change the cost of the same work.
One index is not a verdict. Teams already on Claude should test GPT 6.1 Sol on their own prompts, documents and agent workflows before switching, and check where each provider processes data under the Privacy Act 1988. A short AI strategy review is the cheapest way to route each task to the model that does it best.
A price war fought on efficiency
Anthropic has said Claude Haiku 5.5 is coming in the next few weeks, and OpenAI’s Ultrafast version is days away. With list prices converging, the next round will be decided by tokens per task and speed, which is where our breakdown of the GPT 6 family will need updating as the numbers settle.
