GPT-5.6 Sol vs Terra vs Luna: Pricing, Benchmarks, and Which Tier to Use

GPT-5.6 Sol vs Terra vs Luna: The Short Answer

Last updated November 6, 2026.

GPT-5.6 Sol, Terra, and Luna are one model family sold at three price points. Sol is the flagship and costs $4 input and $20 output per million tokens, a promotional rate that runs until November 30, 2026. Terra is the balanced tier at $2 and $12. Luna is the cost-efficient tier at $0.20 and $1.20, twenty times cheaper than Sol on input. All three share a 1,050,000-token context window, 128,000 max output tokens, and reasoning effort from none to max. OpenAI maps them to the old GPT-5 naming: Sol replaces the unsuffixed model, Terra replaces mini, Luna replaces nano. On the Artificial Analysis Coding Agent Index, Sol at max reasoning scored 80, Terra lands just above Claude Fable 5, and Luna beats Claude Opus 4.8. Pick Terra by default, Sol for the hardest long-running work, and Luna for subagents and bulk calls. Check GPT-6 first: gpt-6.1-sol now costs half of Sol's promo price.

GPT-5.6 Tiers Side by Side

The tiers are not different architectures exposed through different APIs. They share the same endpoints (Chat Completions, Responses, Batch), the same 1.05M window, and the same reasoning controls. OpenAI's model pages describe Sol as the model "for complex professional work," Terra as the one that "balances intelligence and cost," and Luna as "optimized for cost-sensitive workloads."

GPT-5.6 Pricing: Sol, Terra, and Luna

OpenAI has cut GPT-5.6 prices twice since the July 9, 2026 launch. On July 30 it reduced Luna by 80% and Terra by 20%. On August 21 it cut Sol for three months, and the Sol model page now says the $4/$20 promotional rate holds until November 30, 2026, down from $5/$30.

Benchmarks: How Far Apart Are the Tiers?

OpenAI's launch post reports tier results on the Artificial Analysis Coding Agent Index. Sol at max reasoning set a state-of-the-art 80, 2.8 points above Claude Fable 5, using less than half the output tokens. Terra lands just above Fable 5. Luna beats Claude Opus 4.8. OpenAI says Terra and Luna each get there in roughly one-third of the time and at about one-quarter of the estimated cost.

OpenAI did not publish per-tier numbers for every benchmark, so the table leaves gaps rather than filling them from third-party estimates. Notion's co-founder Simon Last is quoted in the same post saying many agents running GPT-5.5 perform just as well on Terra.

Cost per Agent Task

Per-token prices hide the real gap. Take one coding agent task that reads 1M input tokens across its turns and writes 200K output tokens, with every request under 272K.

Output dominates on Sol and Terra once caching kicks in. A Terra run with good cache hits costs about the same as an uncached Sol run on input alone. Luna is cheap enough that ten parallel Luna subagents cost less than one Terra pass.

Which GPT-5.6 Tier to Use

A common split is Terra or Sol as the planner and Luna for everything that reads a lot and decides little. Code search is the biggest of those jobs. Cognition measured that its agent spent 60% of its time searching for code, and WarpGrep moves that work into a separate search subagent so the main model's context holds only the files it needs.

GPT-5.6 vs GPT-6 Pricing

GPT-5.6 is no longer the newest OpenAI family. The same pricing page lists GPT-6 models at half the GPT-5.6 rate or less, so check them before committing to a 5.6 tier.

For full rate cards across providers, see the LLM API pricing guide and OpenAI API pricing.

Frequently Asked Questions