DEV Community

mpoper
mpoper

Posted on Originally published at blog.hefu.hk

GPT-5.6 API Pricing: A Developer's Guide (August 2026)

OpenAI's GPT-5.6 API is officially out of preview and pricing changed on Jul 30, 2026. Here's what developers need to know before building on Sol, Terra, or Luna.

TL;DR

  • Sol: $5/$30 per M input/output tokens
  • Terra: $2/$12 (20% cut from $2.50/$15)
  • Luna: $0.20/$1.20 (80% cut from $1/$6)
  • Cached reads get 90% off; cache writes cost 1.25x input.
  • Inputs over 272K tokens incur 2x input / 1.5x output surcharge.
  • No permanent free tier as of Aug 2026.

The lineup

Model Input Output Vibe
Sol $5/M $30/M Flagship reasoning, Terminal-Bench 2.1 91.9% Ultra
Terra $2/M $12/M Balanced production workhorse
Luna $0.20/M $1.20/M High-throughput, cheap classification/extraction

Why Terra is the sweet spot

Terra now costs 40% of the previous flagship's $5/$30 rate. For agentic loops, RAG pipelines, and multi-step tool use, that's a 60% saving on every token. Luna is 4% of the old flagship price, but you trade reasoning depth.

Cache and long-context gotchas

  • Cached input: Sol $0.50/M, Terra $0.20/M, Luna $0.02/M.
  • Cache writes: Sol $6.25/M, Terra $2.50/M, Luna $0.25/M.
  • Minimum cache lifetime: 30 minutes.
  • Above 272K tokens, input costs 2x and output costs 1.5x. Sol effectively becomes $10/$45 per M on the overage.

Batch discounts and free tier

No separate batch endpoint discount is published for GPT-5.6 as of Aug 2026. There is also no permanent free tier. Preview promo credits were time-boxed; check OpenAI's pricing page for current offers.

Bottom line

GPT-5.6 is the most cost-efficient flagship family OpenAI has shipped — if you watch context length and cache usage. For most production workloads, Terra is the default; Luna is for scale; Sol is for hard reasoning.

Always verify current prices on the official pricing page.

Top comments (0)