GPT-5.4 nano versus GPT-6 Luna

What are we comparing?

GPT-5.4 nano and GPT-6 Luna are two budget models, built for high volumes at a low price, both built by OpenAI. GPT-5.4 nano was released on March 17, 2026, GPT-6 Luna on September 22, 2026.

All figures below come from public sources checked on October 22, 2026: models.dev and OpenRouter for prices and capabilities, Epoch AI for scores.

Which one is more capable?

On our capability index, GPT-5.4 nano scores 71.7 out of 100 and GPT-6 Luna 80.3. GPT-6 Luna leads by 8.6 points.

On the 3 benchmarks reported for both, GPT-5.4 nano wins 0 times and GPT-6 Luna 3:

• GPQA Diamond (PhD-level science questions): GPT-5.4 nano 78.5%, GPT-6 Luna 90.5%
• OTIS Mock AIME (competition mathematics problems): GPT-5.4 nano 87.8%, GPT-6 Luna 98.9%
• FrontierMath (very hard research-level mathematics): GPT-5.4 nano 44.9%, GPT-6 Luna 78.9%

Which one costs less?

GPT-5.4 nano costs $0.20 per million input tokens and $1.25 per million output tokens; GPT-6 Luna costs $0.10 and $0.50.

For a typical agent task (read 20,000 tokens, write 2,000), expect $0.007 with GPT-5.4 nano versus $0.003 with GPT-6 Luna. GPT-6 Luna is 2.2 times cheaper, a gap that adds up quickly over 1,000 tasks: $3.5 difference.

Run the numbers with your own volumes in the agent cost calculator.

Context and capabilities

GPT-6 Luna reads more in one go: 1,050,000 tokens of context (roughly 1,575 pages), versus 400,000 for GPT-5.4 nano.

Reasoning mode: GPT-5.4 nano yes, GPT-6 Luna yes. Tool calling: GPT-5.4 nano yes, GPT-6 Luna yes. Image input: GPT-5.4 nano yes, GPT-6 Luna yes.

Neither publishes its weights: both are used through an API.

Knowledge cutoff: August 31, 2025 for GPT-5.4 nano, May 18, 2026 for GPT-6 Luna.

In short

Our reading of the data:

• Choose GPT-6 Luna for a higher capability index, a lower cost, a larger context window and a more recent release.

This verdict relies on public figures; it does not replace a trial on your own tasks.