Small businesses using AI APIs can leverage the cheaper Luna and Terra models to reduce operational costs while maintaining high performance for automated tasks.
What’s GPT-5.6 and what changed?
OpenAI released the GPT-5.6 family, introducing three distinct model sizes named Luna, Terra, and Sol.
The models are priced per 1M input and output tokens, starting with Luna at $1 and $6. Terra sits in the middle at $2.50 and $15, while Sol costs $5 and $30 per 1M tokens.
OpenAI is aggressively competing on API token pricing for automated workflows.
What’s the evidence behind GPT-5.6?
OpenAI claims GPT-5.6 Sol sets a new high of 53.6 on an evaluation of long-running professional workflows across 55 fields.
This score eclipses Claude Fable 5 by 13.1 points, and even at medium reasoning, Sol beats Fable 5 by 11.4 points at roughly one-quarter the estimated cost. The efficiency extends to the smaller models, with Terra and Luna outperforming Fable 5 at around one-sixteenth the cost.
GPT-5.6 delivers documented cost savings for long-running agentic tasks.
How does GPT-5.6 compare to the alternatives, and what background do small business owners need?
While GPT-5.6 wins on long-running agentic performance, Anthropic’s Claude Fable 5 still dominates SWE-Bench Pro.
On SWE-Bench Pro, Fable 5 scored 80% compared to GPT-5.6 Sol getting 64.6%. OpenAI audited the benchmark and estimates that 30% of the tasks are broken, advising developers to examine results carefully. For comparison, Claude Opus is priced at $5 and $25 per 1M tokens, and Fable 5 is $10 and $50.
Competitors still hold the advantage in complex coding tasks despite GPT-5.6 cost efficiencies.
How does GPT-5.6 affect day-to-day operations for small businesses?
Founders using AI APIs for automated tasks can immediately cut costs by adopting the smaller Luna and Terra models.
New API features include code execution that orchestrates tool calls and the ability to spin up subagents for parallel work. You can now set explicit cache breakpoints to optimize costs, and image requests can bypass automatic resizing. The least expensive run is gpt-5.6-luna at effort none for 0.71 cents, while the most expensive is gpt-5.6-sol at max reasoning for 48.55 cents. You can map these new capabilities against your existing automation stack to find immediate savings.
Switching to Luna or Terra drops API overhead while maintaining high performance.
The sharp hum of the commercial trimmer fades, and the crew supervisor looks at the invoice for the day’s fuel and blade wear. You pay a premium for a heavy-duty mower that burns through complex, overgrown acreage, but for the standard residential lots, running that same machine wastes capital. OpenAI just handed you a smaller, efficient trimmer with GPT-5.6 Luna at $1 per 1M tokens. If your automated workflows are just clearing standard text processing tasks, paying $10 per 1M tokens for a larger model is like pushing a wide-deck mower over a tiny city courtyard. You drop down to Luna or Terra, which outperform competitors at one-sixteenth the cost, and your API overhead stops bleeding into your margins. You get the same clean cut for a fraction of the fuel cost.
What’s the final verdict on GPT-5.6?
GPT-5.6 is a mandatory upgrade for small business owners looking to slash API costs on long-running workflows.
By adopting Terra or Luna at $1 or $2.50 per 1M input tokens, you retain high performance while cutting costs by up to one-sixteenth compared to premium alternatives. You should benchmark your current automated tasks today to see if they fit the Luna or Terra tier.
Optimize your API usage by routing standard tasks to the cheapest capable model.
Source: simonwillison.net
