GPT-6 Sol and Luna Reset the API Cost Curve

GPT-6 Sol and Luna bring 1.05M-token context and sharply lower API prices, widening the gap between premium reasoning and high-volume AI work.

Key takeaways

Sol is priced at $2 input and $10 output per million standard short-context tokens.
Luna is priced at $0.10 input and $0.50 output on the same basis.
Both expose a 1.05-million-token context window, so workload routing now matters more than choosing one default model.

What happened

OpenAI released GPT-6 Sol and Luna on September 22, expanding the GPT-6 family below Astra. Both accept text and image inputs through the Responses and Chat Completions APIs, offer a 1.05-million-token context window and generate up to 128,000 output tokens. OpenAI positions Sol for complex coding and agentic work, while Luna is aimed at focused, high-volume tasks where unit economics matter more than maximum capability.

Why GPT-6 Sol and Luna change the buying decision

The launch is less about a single benchmark than about cost separation. On OpenAI’s standard short-context tier, Sol costs $2 per million input tokens and $10 per million output tokens. Luna costs $0.10 and $0.50. Astra remains $10 and $50. That creates three distinct operating bands inside one family: premium work that can justify Astra, demanding production jobs that fit Sol, and repetitive classification, extraction or support tasks that can move to Luna.

The headline price is not the whole bill

Developers need to match the quoted price to the processing tier and context length they actually use. Cached input, cache writes, long-context prompts, regional processing and faster service modes carry different rates. A long prompt can also increase latency and make a cheap model expensive through sheer token volume. The practical comparison is cost per completed, accepted task—not price per token in isolation.

Routing becomes the product layer

A sensible deployment does not send every request to one model. Teams can route simple, well-specified work to Luna; escalate ambiguous or multi-step work to Sol; and reserve Astra for the hardest reasoning. That architecture needs evaluation sets, confidence thresholds and fallbacks. Without them, low prices can encourage more calls while quietly increasing review work, retries or error handling.

What teams should test now

The first tests should use real tasks and measure accuracy, tool-use success, latency, total tokens, human correction time and escalation rate. Teams should also re-check safety and data-residency settings because model availability does not erase governance requirements. GPT-6 Sol and Luna make intelligence cheaper at the API boundary, but the durable saving will come from choosing the smallest model that reliably completes each job.

The event moves from disclosure through deployment to measurable outcome.

Facts table

Public release	22 September 2026
Sol standard short-context price	$2 input / $10 output per 1M tokens
Luna standard short-context price	$0.10 input / $0.50 output per 1M tokens
Context window	1.05 million tokens
Maximum output	128,000 tokens

Frequently asked questions

How much does GPT-6 Sol cost?

OpenAI lists standard short-context pricing at $2 per million input tokens and $10 per million output tokens.

How much does GPT-6 Luna cost?

OpenAI lists $0.10 per million input tokens and $0.50 per million output tokens on the same tier.

Do both models support images?

Yes. OpenAI says both accept text and image inputs and generate text.

Which model should a business choose?

Use task-level evaluations: Luna for focused high-volume work, Sol for more complex coding and agents, and Astra only when its higher capability changes outcomes.