GPT-6 Astra Sol Luna

Overview

Three GPT-6 SKUs on one ladder. Astra is the most capable ($10 / $50). Sol is the default professional model ($2 / $10). Luna is the efficient high-volume model ($0.10 / $0.50). OpenAI trained Sol and Luna with similar methods to Astra, then cut Sol/Luna API prices 50% vs GPT-5.6 promo rates.

Capabilities

• AstraComplex reasoning, coding, computer use, research, documents. Effort up to xhigh/max. Don’t use it for autocomplete-style volume.
• SolSame $2 / $10 band as Claude Sonnet 5.5. Default for ChatGPT Work / Codex-class tasks when you are in the OpenAI stack.
• LunaMost efficient GPT-6. Free/Go desktop access was called out at launch for Luna; API is `gpt-6-luna`.

Context Window

All three share the >272K input long-context multiplier (2× input and cache, 1.5× output for the full request). Astra at that tier is $20 / $75 class.

• Design prompts to stay under 272K.
• Cache static prefixes—Astra cache reads are $1 / MTok, still cheaper than fresh $10 input.

Pricing

OpenAI standard rates. Batch/Flex 50%; Fast 2×. Long-context column on the pricing page lists higher numbers for the same SKUs.

Astra

$10 / $50 (short) · $20 / $75 (long in/out class)

Sol

$2 / $10 (short) · $4 / $15 (long class)

Luna

$0.10 / $0.50 (short) · $0.20 / $0.75 (long class)

• 50K in + 2K out (short, no cache): Astra ≈ $0.60, Sol ≈ $0.12, Luna ≈ $0.006.
• Tool-specific models may add per-call fees (search, computer use).

OpenAI pricing

How to Use

1. Router: Luna → Sol → AstraCheap first, spend only when quality drops.
2. Pin snapshotsBill against the id in the docs (`gpt-6-astra`, `gpt-6-sol`, `gpt-6-luna` or dated snapshots).
3. Compare to ClaudeSol ≈ Sonnet 5.5 on sticker; Astra ≈ Fable 5.1; Opus 5.5 sits between Sol and Astra at $4 / $20.

Alternatives

• GPT-6 Sol vs LunaTwo-way mid vs low.
• Opus 5.5 vs AstraAnthropic Opus vs Astra.
• Sonnet 5.5Claude $2 / $10 cost page.