GLM-5 vs Kimi K2.5

What are we comparing?

GLM-5 and Kimi K2.5 are two open-weight models that you can install on your own servers, one from Zhipu AI (Z.ai), the other from Moonshot AI. GLM-5 was released on February 12, 2026, Kimi K2.5 on January 27, 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, GLM-5 scores 77.2 out of 100 and Kimi K2.5 78.1. The gap is too small to separate them.

On the 4 benchmarks reported for both, GLM-5 wins 2 times and Kimi K2.5 2:

• GPQA Diamond (PhD-level science questions): GLM-5 87.8%, Kimi K2.5 87.6%
• OTIS Mock AIME (competition mathematics problems): GLM-5 80%, Kimi K2.5 92.2%
• SWE-bench Verified (fixing real software bugs): GLM-5 72.1%, Kimi K2.5 73.8%
• Terminal-Bench (agent tasks in a terminal): GLM-5 52.4%, Kimi K2.5 43.2%

Which one costs less?

GLM-5 costs $0.60 per million input tokens and $2.69 per million output tokens; Kimi K2.5 costs $0.58 and $3.

For a typical agent task (read 20,000 tokens, write 2,000), expect $0.017 with GLM-5 versus $0.018 with Kimi K2.5. The cost is almost the same.

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

Under the hood

Kimi K2.5 reads more in one go: 262,144 tokens of context (roughly 393 pages), versus 204,800 for GLM-5.

Reasoning mode: GLM-5 yes, Kimi K2.5 yes. Tool calling: GLM-5 yes, Kimi K2.5 yes. Image input: GLM-5 no, Kimi K2.5 yes.

Both publish their weights: you can host them yourself.

In short

Our reading of the data:

• Choose Kimi K2.5 for a larger context window and image input.

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