TLDR
OpenAI dropped GPT-6 Sol and GPT-6 Luna on 22 September 2026, cutting API prices 50 per cent against its own GPT-5.6 rates and landing Luna at USD 0.10 per million input tokens, the cheapest frontier-class option announced that day.
The timing was not accidental
OpenAI listed GPT-6 Sol and GPT-6 Luna in its API changelog on 22 September 2026, within minutes of Anthropic publishing Claude Opus 5.5.[1] The sequencing was tight enough to read as deliberate: drop matching frontier models the moment a competitor's launch goes live, and you own the price comparison before anyone writes one.
GPT-6 Sol is priced at USD 2 per million input tokens and USD 10 per million output, while GPT-6 Luna sits at USD 0.10 input and USD 0.50 output.[1] OpenAI said those figures represent a 50 per cent reduction against GPT-5.6 promotional pricing.[2] Anthropic's Opus 5.5, released the same morning, is priced at USD 4 per million input tokens and USD 20 per million output, twice the Sol rate on both sides of the ledger.[3]
What the models do
OpenAI positioned Sol as its high-capability tier and Luna as the cost-optimised workhorse. Neither company has published independent benchmark audits at time of writing, so the performance claims on each developer documentation page remain self-reported.
The capability case for Claude Opus 5.5 rests on agentic, multi-step engineering work. Sean Heintz, a staff software developer at Clio, said he handed Opus 5.5 a large engineering task spanning six repositories, let it run unattended overnight for over 18 hours, and found the model had defined how services talk to each other with code comments that were short and useful rather than long and prose-heavy, requiring minimal reworking compared with Opus 5.[3] Mario Rodriguez, Chief Product Officer at GitHub, said that in testing across GitHub Copilot CLI and VS Code, Claude Opus 5.5 solved more terminal tasks than Opus 5 in less than half the steps.[3]
Rodriguez said developers want agents capable of taking on real software work and finishing it, and that Opus 5.5 makes developers' bigger projects more achievable, beyond simply speeding up individual tasks.[3]
What Australian budgets see
For Australian agencies and publishers running high-volume coding pipelines or content generation at scale, the token price is the bill. OpenAI's 50 per cent price reduction against its own GPT-5.6 promotional rates means the same monthly API spend now buys twice the throughput on Sol or Luna.[2] Luna at USD 0.10 per million input tokens sits well below any comparable frontier-class option announced on 22 September 2026, making it the practical default for batch data jobs where output quality tolerates a step down from Sol.
Two rounds of significant price reductions in quick succession have compressed the gap between frontier-model capability and commodity-tier pricing faster than most enterprise procurement cycles anticipated. Australian businesses that locked in annual AI contracts at 2025 rates should check those terms against the current changelog before their next renewal.
KEY TAKEAWAYS
SOURCES & CITATIONS
FREQUENTLY ASKED QUESTIONS
How do GPT-6 Sol and Luna prices compare with Claude Opus 5.5?
When did OpenAI release GPT-6 Sol and Luna?
What are the main capability claims for Claude Opus 5.5?
Alex Mercer writes about technology, energy and infrastructure. He likes the physical end of the story: the plants, the grids and the machines that everything else depends on.
