Compare GPT-6 Astra, Sol and Luna pricing per 1M tokens and estimate your monthly API spend in seconds. Free, no signup.
// 01 · Model
// 02 · Service tier
Flex and Batch bill at 50% off Standard rates — same tokens, half the price, lower priority. Fast mode (priority throughput) has no public multiplier, so it's excluded rather than guessed.
// 03 · Workload
Standard tier, per 1M tokens. Flex and Batch bill at 50% off these rates. Snapshot — providers change prices, always verify on the official pricing page.
| Model | Input / 1M | Cached input / 1M | Output / 1M | Best for |
|---|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.01 | $0.50 | High-volume, cost-efficient workloads |
| GPT-6 Sol | $2.00 | $0.20 | $10.00 | Flagship quality, complex reasoning |
| GPT-6 Astra | $10.00 | $1.00 | $50.00 | Top-tier quality, enterprise workloads |
Prices last verified: September 26, 2026 · Source: OpenAI platform pricing (verify at platform.openai.com/docs/pricing)
Same workload, very different bills. Luna wins on price by a mile; the flagships compete on quality.
| Model | Input / 1M | Output / 1M | Relative cost |
|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.50 | Baseline — cheapest |
| GPT-6 Sol | $2.00 | $10.00 | ~20× Luna |
| GPT-6 Astra | $10.00 | $50.00 | ~100× Luna |
| Claude Opus 5.5 | $4.00 | $20.00 | ~40× Luna |
Rule of thumb: if Luna's quality clears your bar, nothing else comes close on price. Pay for Sol/Opus only when the task genuinely needs frontier reasoning.
OpenAI bills GPT-6 by the token, split into input and output. Input is everything you send — system prompt, conversation history, the user message, tool definitions. Output is everything the model generates, including hidden reasoning tokens. Output is priced higher because generation is the expensive part.
A RAG chatbot with long retrieved context might run 10:1 input-to-output. A code generator might run 1:3. Two apps with the same request count can differ 10× in cost. Before optimizing the model choice, measure your real token split — most developers overestimate output and underestimate how much history they resend every turn.
If your system prompt and context repeat across requests, cached input pricing (a fraction of standard input rates) applies to the repeated prefix. Apps with stable system prompts routinely cut input spend 50–80% with caching. In the calculator above, move the cached-input slider to see the effect on your workload.
The same model has three price points. Standard is full price at full speed. Flex costs about half but your requests queue behind Standard traffic — fine for background jobs, bad for user-facing latency. Batch also runs ~50% off with up to 24-hour turnaround for non-urgent bulk work. Toggle the tier in the calculator to see the difference on your workload.
Luna is the price-performance pick: classification, extraction, simple chat, bulk processing. Sol earns its premium on hard reasoning — math, complex code, agentic multi-step tasks where a weaker model's mistakes cost more than the price gap. Astra is the top shelf: 5× Sol's price for the highest quality bar, aimed at enterprise workloads where only the best output will do. Prototype on Sol, then test whether Luna holds quality on your evals; most teams are surprised how often it does.
Quick answers to the pricing questions developers actually search for.