GPT-6 Luna is OpenAI’s lowest-cost GPT-6 model. At $0.10 per million input tokens and $0.50 per million output tokens under Standard processing, it targets focused, high-volume tasks rather than the hardest autonomous work.
The API model ID is gpt-6-luna. It supports text and image input, text output, a 1,050,000-token context window and up to 128,000 output tokens. Those limits match GPT-6 Sol, but equal context size does not mean equal task capability.
GPT-6 Luna pricing
| Processing | Input | Cached input | Cache write | Output |
|---|---|---|---|---|
| Standard | $0.10 | $0.01 | $0.125 | $0.50 |
| Batch / Flex | $0.05 | $0.005 | $0.0625 | $0.25 |
| Fast | $0.20 | $0.02 | $0.25 | $1.00 |
Prices are per million tokens. Above 272,000 input tokens, the entire request uses $0.20/M input, $0.02/M cached input, $0.25/M cache writes and $0.75/M output. Avoid treating the 1.05M window as free capacity.
Standard is the documented path for EU data residency. Regional processing adds 10%, while Fast does not support EU data residency.
API and reasoning support
Luna supports Responses, Chat Completions and Batch. Its reasoning levels run from none through max, with medium as the default. Chat Completions function calling is limited to reasoning none; use Responses for richer agent workflows.
The most natural Luna workloads are classification, extraction, routing, formatting, short transformations and inexpensive first-pass processing. Validate accuracy and escalation rules before using it for consequential decisions or long autonomous workflows.
Luna in GitHub Copilot
GitHub is gradually rolling Luna out to Copilot Pro, Pro+, Max, Business and Enterprise. That is broader plan coverage than Sol, which starts at Pro+. The model appears across GitHub’s supported IDE, CLI, cloud-agent, app, web and mobile surfaces, subject to rollout and administrator model policy.
Copilot usage-based billing is distinct from OpenAI API token billing. Check the selected-model charge in Copilot rather than applying direct API rates to a Copilot session.
Luna versus Sol and Astra
Use Luna when volume and latency matter and the work is narrow enough to evaluate automatically. Move to Sol when tasks require stronger multi-step coding or tool judgment. Reserve Astra for cases where its added capability produces a measurable reduction in failures or human intervention.
An economical production pattern is Luna first, with a guarded escalation to Sol for low-confidence or failed cases. Instrument the route, retries and final task success through AI Operations rather than assuming a cheaper first call always means a cheaper workflow.
Verdict
GPT-6 Luna makes the GPT-6 family viable for high-volume processing, but its value depends on disciplined routing. It is the cheapest GPT-6 option, not a universal replacement for Sol or Astra.
See GPT-6 Sol vs GPT-6 Luna for exact workload calculations and a practical escalation policy.