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

OpenAI launched GPT-5.6 with three tiers instead of one model. Sol is the flagship. Terra is the workhorse. Luna is the budget option. They share the same architecture but differ in size, speed, and price.

Here's the bottom line: Sol beats GPT-5.5 at the same price. Terra delivers GPT-5.5-level quality at half the cost. Luna gives you solid performance for $1/$6 per million tokens. All three have 1.05M context windows and 128K max output.

The Three Tiers at a Glance

The quality jump from Luna to Sol is 15 points. That's significant. But for many tasks, the difference between 50 and 65 doesn't matter. Emails, simple coding, summarization, content drafting. Luna handles all of these without issues.

GPT-5.6 Sol: The Flagship

Sol sits at the top of the OpenAI lineup with a quality score of 65. That puts it second overall behind Claude Fable 5 (68) and ahead of Opus 4.8 (61). The math score of 91 is the highest of any GPT model to date.

At $5/$30, it's the same price as the old GPT-5.5. You're getting a straight upgrade for free.

Sol supports five reasoning effort levels. At none or low, it behaves like a fast production model. At high or xhigh, it spends more time thinking through complex problems. This flexibility means one model can serve multiple roles in your stack.

When to use Sol: Architecture decisions. Complex debugging. Security-sensitive code review. Problems that stump cheaper models. Mathematical proofs and reasoning.

GPT-5.6 Terra: The Daily Driver

Terra is where most developers should start. It matches GPT-5.5 on quality (both score 57 on the intelligence index) at half the output cost: $15 vs $30 per million tokens.

The coding score of 55 puts it on par with Gemini 3.1 Pro and ahead of Claude Sonnet 4.6. The 75 tok/s speed is comfortable for interactive use.

Terra replaced GPT-5.4 as the go-to premium model. Same $2.50/$15 price point, better scores.

When to use Terra: Daily coding. Feature implementation. Data analysis. Content writing. Most work that previously went to GPT-5.4 or GPT-5.5.

GPT-5.6 Luna: The Budget Option

Luna is the new entry in OpenAI's lineup and it fills a gap. At $1/$6 per million tokens, it's 5x cheaper than Terra on output and 10x cheaper than Sol. But with a quality score of 50, it outperforms models that cost 2-3x more.

The speed is the standout. Luna runs at 150 tok/s, making it the fastest model in the GPT family. That speed makes it ideal for agent subagent work where you need quick responses to many small tasks.

The 1.05M context window at this price is unusual. Most models under $10/M blended cost have smaller windows. Luna gives you full context at a fraction of the price.

When to use Luna: Agent subagent tasks. Email drafts. Summarization. Code explanation. Simple edits. Anything where speed and cost matter more than peak quality.

How They Compare to Claude and Gemini

Sol vs Fable 5: Sol is cheaper ($5/$30 vs $10/$50), faster (85 vs 20 tok/s), but lower quality (65 vs 68). For most tasks, Sol is the better value. Fable 5 wins only on the hardest problems.

Terra vs Sonnet 5: Close competition. Sonnet 5 scores slightly higher (58 vs 57) at a lower intro price ($2/$10 vs $2.50/$15). After August 31, Sonnet 5 moves to $3/$15 and Terra becomes the value pick. Terra is faster (75 vs 55 tok/s).

Luna vs Gemini 3 Flash: Luna costs more ($1/$6 vs $0.075/$0.30) but scores significantly higher (50 vs 46). Flash wins on raw price. Luna wins on quality per dollar for tasks that need more intelligence than Flash can deliver.

Real-World Costs

Monthly estimates for 500 tasks at 20,000 tokens average:

Luna at $35/month covers most professional use cases. For teams processing high volumes, that's roughly $1.15 per day.

Best Configuration for July 2026

A recommended two-tier setup using the GPT-5.6 family:

Primary model: GPT-5.6 Terra ($2.50/$15) Handles complex tasks, code review, analysis, writing.

Subagent model: GPT-5.6 Luna ($1/$6) Handles delegated tasks, simple edits, lookups, summarization.

Estimated monthly cost: $40-80 depending on volume.

If you need maximum quality, swap Terra for Sol as primary and keep Luna for subagents. Monthly cost rises to $80-160.

Or mix providers: Claude Sonnet 5 as primary (better quality per dollar during intro pricing) with Luna as subagent.

FAQ

Is GPT-5.6 Sol better than GPT-5.5?

Yes. Sol scores 65 on the intelligence index vs. GPT-5.5's 60. Same price ($5/$30). It's a straight upgrade. There's no reason to use GPT-5.5 if Sol is available.

Which GPT-5.6 model should I use for coding?

Start with Terra ($2.50/$15). It handles 90% of coding tasks well. Use Sol for complex debugging or architecture work. Use Luna for simple edits and code explanation.

Is GPT-5.6 Luna good enough for production?

For many use cases, yes. Luna scores 50 on the intelligence index with reasoning capabilities. It handles email, summarization, simple coding, and agent subagent work at $1/$6 per million tokens. For complex analysis or frontier-quality output, step up to Terra or Sol.

How does GPT-5.6 compare to Claude Fable 5?

Claude Fable 5 (score 68) beats GPT-5.6 Sol (score 65) on raw quality, but costs twice as much ($10/$50 vs $5/$30) and runs much slower (20 vs 85 tok/s). Sol is the better value for most tasks. Fable 5 is for problems where the extra 3 quality points matter.

When will GPT-5.6 be available on OpenRouter?

GPT-5.6 Sol, Terra, and Luna are expected to be available on OpenRouter shortly after GA launch. Check OpenRouter's model page for current availability and pricing.

Pricing from OpenAI's announcement. Benchmark scores from Artificial Analysis. Updated July 2026. Compare all models on our Benchmark Dashboard or get personalized picks with the Model Selector.