# GPT-5.6 Luna on the Cotera agent benchmark

> The $1-per-million tier went 5/5 for $0.30 — the second-cheapest perfect score in the matrix.

Source: https://cotera.co/benchmarks/models/azure-gpt-5.6-luna

---

OpenAI's GPT-5.6 Luna — the fast/affordable tier of the July 2026 GPT-5.6 preview at $1/$6 per 1M tokens. The spiritual successor to GPT-5 Mini's price point with a 1.05M context window.

**Score:** 5/5 benchmarks passed · **Cost across the matrix:** $0.296

## Strengths
- Five for five at $0.30 total — only Mistral Large 3 ($0.27) has done the matrix perfectly for less.
- Now holds the single cheapest pass in the matrix: $0.021 on the Apollo pricing scrape, edging out Mistral.
- CX in two tool calls and $0.06 — the benchmark that cost its flagship sibling the perfect score, done for a quarter of the price.

## Watch-outs
- It reads more than it needs to: on Sales it scraped the full Crunchbase page before calling the Crunchbase tool, making its $0.089 Sales run the most expensive of the three tiers.
- Preview model at preview pricing — the $1/$6 rate is the cheap-tier hook and may move at GA.
- No data yet on long-horizon behavior; the 1.05M window invites big-context jobs this suite does not test.