Kimi K2.5 vs GLM 5
Moonshot AIZ.aiboth via one key, provider price + 3%
Kimi K2.5 (Moonshot AI, released 2026-01-27) and GLM 5 (Z.ai, released 2026-02-11) are both recent agent-focused models, but they differ in modality, context window, and price. Kimi K2.5 handles text and image input; GLM 5 is text-only. Kimi K2.5 has a 262,144-token context window versus GLM 5's 202,752 — a 1.29x ratio in Kimi's favor. Both run on OpenKey under a single API key with a flat 3% fee on top of provider list pricing.
Spec vs spec
| Spec | Kimi K2.5 | GLM 5 |
|---|---|---|
| Context window | 262K | 203K |
| Input modalities | text, image | text |
| Output modalities | text | text |
| Released | Jan 27, 2026 | Feb 11, 2026 |
| Reasoning | optional | optional |
Pricing
Per 1M tokens. Provider price plus the flat 3% fee — the sum is what you pay.
moonshotai/kimi-k2.5
Input · 1M tokens
$0.375 + 3%$0.386
Output · 1M tokens
$2.02 + 3%$2.09
FEE — FLAT, EVERY MODEL3%
z-ai/glm-5
Input · 1M tokens
$0.600 + 3%$0.618
Output · 1M tokens
$1.92 + 3%$1.98
Cache read · 1M tokens
$0.120 + 3%$0.124
FEE — FLAT, EVERY MODEL3%
One workload, priced on both
10M input + 2M output tokens at each model's price, flat 3% fee included.
moonshotai/kimi-k2.5Cheaper
$8.03
$7.80 provider + 3%
z-ai/glm-5
$10.14
$9.84 provider + 3%
Benchmarks
Design Arena categories where both models have results. Higher Elo and lower rank win.
| Kimi K2.5 | GLM 5 | |||
|---|---|---|---|---|
| Category | Elo | Rank | Elo | Rank |
| 3D | 1286 | #22 | 1307 | #15 |
| androidnative | 1132 | #17 | 1244 | #6 |
| ASCII art | 1214 | #17 | 1192 | #26 |
| Code | 1286 | #20 | 1295 | #16 |
| Data viz | 1270 | #21 | 1269 | #22 |
| Full-stack | 1182 | #14 | 1190 | #13 |
| Game dev | 1272 | #23 | 1299 | #15 |
| godotgamedev | 1254 | #2 | 1237 | #3 |
| Mobile apps | 1186 | #20 | 1222 | #10 |
| SVG | 1210 | #25 | 1225 | #19 |
| UI components | 1290 | #19 | 1287 | #21 |
| Websites | 1291 | #16 | 1290 | #18 |
Head-to-head preference voting. How we filter and rank
Pricing math
Kimi K2.5 costs $0.375/M input and $2.025/M output from the provider; on OpenKey that's $0.375 × 1.03 = $0.38625/M input and $2.025 × 1.03 = $2.08575/M output. GLM 5 runs $0.60/M input and $1.92/M output provider-side, or $0.60 × 1.03 = $0.618/M input and $1.92 × 1.03 = $1.9776/M output on OpenKey. GLM 5 also offers a $0.12/M cache-read rate, which Kimi K2.5 doesn't list.
For a 10M-input / 2M-output workload, Kimi K2.5 comes out to $7.80 total, GLM 5 to $9.84. That's a $2.04 gap per run, and it compounds fast if you're doing this at scale — Kimi K2.5's input price is 0.62x GLM 5's, meaning it's the cheaper choice by a wide margin on prompt-heavy workloads specifically.
Agent and coding benchmarks
On Design Arena's agent categories, Kimi K2.5 ranks #2 in Godot game dev (elo 1254, 59.5% win rate) and #14 in fullstack (elo 1182, 54.2% win rate). GLM 5 ranks #3 in Godot (elo 1237, 54.8% win rate) and #6 in Android-native (elo 1244, 62% win rate) — its strongest category by rank. Kimi K2.5's Android-native rank is #17 (elo 1132, 57.8% win rate), notably behind GLM 5 here.
In the broader model categories, GLM 5 edges ahead in gamedev (#15, elo 1299) versus Kimi K2.5 (#23, elo 1272), and in 3D (#15, elo 1307 vs Kimi's #22, elo 1286). Kimi K2.5 takes website (#16, elo 1291 vs GLM 5's #18, elo 1290) by a thin margin. Neither model dominates across the board — pick based on which category matches your workload.
Context and modality
Kimi K2.5 supports 262,144 tokens of context against GLM 5's 202,752 — about 1.29x more room, useful if you're feeding large codebases or long documents into a single call. Kimi K2.5 also accepts image input alongside text, which GLM 5 does not support at all (text-to-text only). If your pipeline needs to parse screenshots, diagrams, or scanned docs, GLM 5 is off the table and Kimi K2.5 is the only option between the two.
When to pick each
Pick Kimi K2.5 when cost matters, when you need image input, or when your agent work leans toward Godot game dev or website generation — categories where it ranks higher. Pick GLM 5 when your workload is Android-native agent tasks specifically (rank #6 vs #17), or when cache-read pricing at $0.12/M provider-side matters for repeated-context calls. Both models are new enough (January and February 2026 releases) that neither has an established track record beyond these benchmark snapshots, so test on your own workload before committing at scale.
Which model for which job
| Use case | Pick | Why |
|---|---|---|
| Cost-sensitive high-volume text generation | Kimi K2.5 | $7.80 vs $9.84 for a 10M-in/2M-out workload |
| Image + text input tasks | Kimi K2.5 | GLM 5 is text-only, no image modality support |
| Android-native agent workflows | GLM 5 | Ranks #6 (elo 1244) vs Kimi K2.5's #17 (elo 1132) |
| Godot game development agents | Kimi K2.5 | Ranks #2 (elo 1254) vs GLM 5's #3 (elo 1237) |
| Long-document or large-codebase context | Kimi K2.5 | 262,144 tokens vs GLM 5's 202,752, a 1.29x larger window |
| Repeated-context calls needing cache pricing | GLM 5 | Lists a $0.12/M cache-read rate; Kimi K2.5 has none listed |
Questions
- Which model is cheaper for a typical workload?
- Kimi K2.5, by a clear margin. A 10M-input/2M-output run costs $7.80 on Kimi K2.5 versus $9.84 on GLM 5 — a $2.04 difference per run, driven mostly by Kimi K2.5's input price being 0.62x GLM 5's.
- Can either model process images?
- Only Kimi K2.5. It supports text and image input with text output. GLM 5 is text-to-text only, so if your pipeline needs to read screenshots or diagrams, GLM 5 isn't an option.
- Which has the bigger context window?
- Kimi K2.5, at 262,144 tokens versus GLM 5's 202,752 tokens — a 1.29x ratio. That extra room matters if you're passing large codebases or long documents in a single call.
- Which model wins more Design Arena agent categories?
- It splits. Kimi K2.5 ranks #2 in Godot game dev (elo 1254) and #14 in fullstack (elo 1182). GLM 5 ranks #6 in Android-native (elo 1244) and #3 in Godot (elo 1237) — GLM 5's Android-native rank is its clearest advantage.