Dramatically reduces API costs for businesses relying on AI for text generation, summarization, or customer service automation.
What is GPT-5.6 Luna and what changed?
OpenAI cut GPT-5.6 Luna’s API price by 80%, dropping input to $0.20 per million tokens and output to $1.20 per million tokens.
GPT-5.6 Terra also received a 20% reduction in the same announcement, which applies immediately for production API usage.
These updated rates reset the cost structure for any business running text generation, summarization, or customer service automation on OpenAI’s API.
Simon Willison verified the announcement and switched his agent.datasette.io demo site from Gemini 3.1 Flash-Lite to Luna the same day, which confirms the model is production-ready.
Willison noted that the Luna price drop completely changes the landscape with respect to lower priced models, since Luna is now cheaper than Google’s Gemini 3.1 Flash-Lite on output.
GPT-5.6 Luna is now the cheapest premium LLM on the market for output-heavy workloads.
What is the evidence behind GPT-5.6 Luna?
AI expert Simon Willison verified the new $0.20 and $1.20 rates and switched his own demo site to Luna the same day.
OpenAI credits GPT-5.6 Sol with optimizing the model’s forward pass, which reduced end-to-end serving costs by 20%.
GPT-5.6 Sol autonomously rewrote production kernels in Triton and Gluon, two open-source GPU programming languages maintained by OpenAI, to avoid inefficient data layouts and synchronization issues.
The optimization found work that could be precomputed, avoided, or parallelized, which is what made the 80% price drop sustainable rather than a temporary promotion.
OpenAI trained GPT-5.6 to be effective at writing and improving kernels in Triton and Gluon, which is why GPT-5.6 Sol could autonomously rewrite the production kernels that execute the mathematical operations making up the model.
Verified engineering optimizations directly enabled this 80% price cut.
How does GPT-5.6 Luna compare to the alternatives, and what background do small business owners need?
GPT-5.6 Luna now significantly undercuts Anthropic on API pricing for both input and output tokens.
Luna’s $0.20 input is 1/5th of Claude Haiku 4.5’s $1.00 input rate, and Luna’s $1.20 output undercuts Haiku’s $5.00 by roughly 4x.
Anthropic’s cheapest current model is Claude Haiku 4.5 at $1 and $5, and Luna previously cost the same on input, which means the gap is new and structural.
Against Google’s Gemini 3.1 Flash-Lite at $0.025 input and $1.50 output, Luna’s $1.20 output wins, but Luna’s $0.20 input costs more than Gemini’s $0.025 input.
The right routing depends on whether your workload is input-heavy or output-heavy, since Gemini wins on input cost while Luna wins on output cost.
Small business owners should route output-heavy workloads to Luna and input-heavy workloads to Gemini 3.1 Flash-Lite.
How does GPT-5.6 Luna affect day-to-day operations for small businesses?
Small business owners relying on AI-powered customer support chat tools will see immediate margin improvements on every routed conversation.
Automated customer service pipelines that process 10 million input tokens a month now cost $2 on Luna instead of $10 on Claude Haiku 4.5.
Founders running heavy summarization or generation workflows should audit their current API usage and switch their routing to Luna for output-heavy calls.
The price reduction applies immediately for production API usage, which means there is no preview period or waitlist to navigate.
Willison’s same-day switch of his demo site from Gemini 3.1 Flash-Lite to Luna confirms the model handles real agent workloads, not just benchmark tests.
Founders must update their API routing today to capture the savings.
The API dashboard refreshes at the close of the billing cycle, and the token spend line item sits unchanged for the third straight month. The team has not renegotiated because swapping models always loses the sprint to the next fire, and the invoice keeps clearing.
OpenAI just dropped GPT-5.6 Luna to $0.20 per million input tokens, which is 1/5th of what Claude Haiku 4.5 charges for the same workload. A support pipeline processing 10 million input tokens a month now costs $2 instead of $10, and the savings land the moment you flip the routing.
The model swap is a routing change, not a rebuild, and the next invoice confirms whether the 80% drop was real. Founders who let this slide another quarter are lighting margin on fire while competitors lock in the new floor.
What is the final verdict on GPT-5.6 Luna?
OpenAI fundamentally reset the pricing floor for capable LLM inference with this 80% cut.
At $0.20 per million input tokens, GPT-5.6 Luna undercuts Claude Haiku 4.5 by 5x on input cost and 4x on output cost.
The 20% internal serving cost reduction validated by GPT-5.6 Sol made this price sustainable, and Simon Willison’s same-day switch confirms the model is production-ready.
Small business owners running automated text pipelines should switch immediately to lock in the margin expansion before competitors catch up.
The pricing change is permanent, not promotional, since it is backed by real engineering optimizations to the inference pipeline.
GPT-5.6 Luna delivers premium AI capability at a fraction of the previous cost.
Source: simonwillison.net
