GPT-6 Luna vs GPT-6.1 Sol: API cost versus reasoning

By TechCompare · Updated · How we check

Luna cuts base input costs by 95% and output costs by 95%. Use Sol where your evaluations show that better completion quality pays for its higher token rates.

OpenAI: GPT-6 Luna

OpenAI's own price

Input
$0.10
Output
$0.50
Cached
$0.01

per 1M tokens

Batch availableFlex available

Price checked 2026-11-19

OpenAI: GPT-6.1 Sol

OpenAI's own price

Input
$2
Output
$10
Cached
$0.10

per 1M tokens

Batch availableFlex available

Price checked 2026-11-19

A newer version is out: GPT-6.2 Sol

Cost Comparison

Based on 100,000 input tokens (50% cached), 5,000 output tokens, and 100 requests.Prices come live from OpenRouter at each maker's own price where it's listed. Temporary deals are left out.

Loading model prices

GPT-6 Luna's base input rate is about 20 times lower than GPT-6.1 Sol's, and its output rate about 20 times lower. That saving adds up over repeated requests, but the useful comparison is the cost of getting an acceptable result. A cheap call that needs several corrections can erase part of its advantage. This article compares equal token workloads first, then explains how to test whether Sol's reasoning is worth the extra spend on your own tasks.

Compare Luna's base token rates with Sol's, then the cost of a successful result.

Side-by-side specs

SpecGPT-6 LunaGPT-6.1 Sol
Rates
Standard input per 1M tokens
$0.10 (better on this spec)
$2.00
Standard output per 1M tokens
$0.50 (better on this spec)
$10.00
Cache reads per 1M tokensCached input is text the provider has already processed. It costs less than fresh input.
$0.01 (better on this spec)
$0.10
Example costs
20K input + 2K output, no cache hits
$0.003 (better on this spec)
$0.06
20K input + 2K output, 50% cached input
$0.0021 (better on this spec)
$0.041
100K input + 5K output, 50% cached input
$0.008 (better on this spec)
$0.155
300K input + 5K output, no cache hits
$0.06375 (better on this spec)
$1.275
Limits and extras
Input / output on long prompts, per 1MSome models charge a higher price for the whole request once the prompt passes a set length.
$0.20 / $0.75 above 272,000 input
$4.00 / $15.00 above 272,000 input
Context window / maximum output
1,050,000 / 128,000 tokens
1,050,000 / 128,000 tokens
Batch API input / output per 1MBatch jobs return results later, at a lower price than live calls.
$0.05 / $0.25
$1 / $5

How they differ

Standard input costs $0.10 per million tokens on Luna and $2 on Sol. Output costs $0.50 and $10, while cache reads cost $0.01 and $0.10. A request with 20,000 input and 2,000 output tokens costs $0.003 on Luna and $0.06 on Sol without cache hits.

Reading half the input from cache changes those totals to $0.0021 and $0.041. Sol discounts cached input more heavily, so caching helps it more. Both models have a 1,050,000-token context window. Both charge a full-request surcharge on long prompts, above 272,000 input tokens on Luna and 272,000 on Sol. At 300,000 input and 5,000 output tokens without cache hits, the estimates are $0.06375 and $1.275.

These figures exclude cache writes and tool fees. The embedded chart uses live OpenRouter rates for its displayed workload.

Verdict

Test a representative task set before choosing one model for all traffic. Start with tasks that have clear success checks, such as valid extracted fields, passing code checks, or a correct calculation. Record total billed usage, corrections, and review time for each accepted result. Route routine work to Luna if it meets your standard, then use Sol for cases where measured improvements justify the cost. Price alone doesn't establish which model reasons better on a particular problem.

Which should you pick?

Choose GPT-6 Luna

Choose Luna when focused requests pass your quality checks and you expect enough volume for small per-call savings to matter. Try it on classification, extraction, and short transformations using examples from your application. Keep outputs concise where the task allows it, and track failures so a low token bill doesn't hide extra manual work or repeated attempts.

Official Luna model reference

Choose GPT-6.1 Sol

Choose Sol when your task set shows that it resolves difficult cases with fewer corrections or less review. Examples to test include changes across several code files, ambiguous instructions, and analysis that must satisfy several constraints. Define that improvement before spending more. A higher reasoning setting can also change output usage, so compare completed tasks using billed tokens rather than equal settings alone.

Official Sol model reference

Frequently asked questions

What would 100,000 requests cost on Luna and Sol?
At 20,000 input and 2,000 output tokens per call without cache hits, the token estimates are $300 on Luna and $6,000 on Sol. If half the input is read from cache, they become $210 and $4,100. For the same uncached workload, the Batch API brings the estimates to $150 and $3,000. Cache writes, tools, and other processing charges are additional.
Is Sol worth paying more for reasoning?
Only if it completes more of your tasks successfully in your own evaluations. Compare both models on the same tasks and score correctness, instruction following, extra calls, and review time. Both support reasoning. Include billed reasoning output in your budget and pay for Sol where its measured improvement matters to the result.
How does caching change the gap between Luna and Sol?
The base input rates have a 20-fold gap and the output rates a 20-fold gap, while cache reads have a 10-fold gap. With 20,000 input tokens, half cached, and 2,000 output tokens, Sol's estimate is $0.041 and Luna's is $0.0021. The ratio depends on the input, output, and cache mix. Creating a cache also costs money, so include writes when estimating a fresh or frequently changing prompt.
Can I send easy tasks to Luna and hard tasks to Sol?
Yes, if you have a way to identify cases that need another model. For example, routing 90% of the half-cached 20K-input, 2K-output workload directly to Luna and 10% directly to Sol costs $189 on Luna plus $410 on Sol over 100,000 calls. That assumes one call per task. A workflow that tries Luna first and then calls Sol needs to include both calls for escalated tasks.