GPT-5.6 Luna vs Kimi K3 benchmark

Side-by-Side Benchmark Matrix

Recommended Workload Routing

Frequently Asked Questions

Which is better overall, GPT-5.6 Luna or Kimi K3?

Kimi K3 has the higher preference Elo in our latest snapshot (1,528). “Better” still depends on coding, price, and latency — see the table.

Which is better at coding, GPT-5.6 Luna or Kimi K3?

Kimi K3 leads SWE-bench at 73.8% vs 61.8%. SWE-bench is one harness, not your repo.

Which is cheaper to run in production?

GPT-5.6 Luna output tokens are $1.2/1M versus $2.5/1M. Input prices and retry rates still move the real bill.

Which is faster, GPT-5.6 Luna or Kimi K3?

GPT-5.6 Luna has the lower time-to-first-token (88 ms vs 290 ms). Tokens/sec is a separate column if you care about long completions.

Which has the larger context window?

GPT-5.6 Luna accepts 1M tokens versus 262k.

Can I self-host either model?

Neither model is marked open-weights here. You are comparing hosted APIs.

Are these scores from CompareLLM’s own evals?

No. V1 aggregates public leaderboards (and optional first-party latency pings). Each cell names its source and date. Read /methodology.

What does preference Elo mean on this page?

Preference Elo is a crowd vote from LMArena / Arena. People see two hidden answers and pick the one they like more. The model that wins more often gets a higher Elo. That means people preferred it — not that it passed a school test. It is not SWE-bench, not accuracy, and not a number we invent. GPT-5.6 Luna and Kimi K3 Elo cells are dated snapshots from the named source — we do not compute Arena ourselves.