Kimi K3 vs GLM-5.2
Kimi K3 wins on Overall, Coding, Multimodal. Every numeric field is compared below, with a worked monthly-cost example and a pick rule for each use case.
Verdict at a glance
Choose Kimi K3 if you…
- Best-in-class long text
- Open-source and self-hostable
- Strong codebase understanding
- Fair price
- Best for: Long-document reading, Codebase analysis, Chinese
Choose GLM-5.2 if you…
- Strong open-source ecosystem
- Good Chinese
- Mature enterprise services
- Best for: Chinese, Open source, Coding
Head-to-head aggregate scores
Scores are 0–100, aggregated from public benchmark information and independently weighted across three leaderboards. Rank is out of 22 tracked models.
Aggregated from public sources and independently weighted; methodology on the Terms page. Scores within 3 points are treated as statistically tied.
Specs & pricing — every field side by side
List API prices in USD per 1M tokens. The highlighted cell is the stronger value/capability on that row.
| Dimension | Kimi K3 | GLM-5.2 | Verdict |
|---|---|---|---|
| Vendor | Moonshot AI (CN) | Zhipu AI (CN) | Different vendors |
| Released | 2026.06 | 2026.06 | GLM-5.2 is newer |
| Overall (rank) | 77 · #8 | 73 · #10 | Kimi K3 +4 |
| Coding | 71 · #11 | 66 · #15 | Kimi K3 +5 |
| Multimodal | 81 · #10 | 79 · #12 | Kimi K3 +2 |
| Context window | 256K | 128K | Kimi K3 larger |
| Max output | 64K | 64K | Tie |
| Effective-context | 99 | 88 | Kimi K3 more reliable |
| Input $/1M | $3 | $1.4 | GLM-5.2 cheaper |
| Output $/1M | $15 | $4.4 | GLM-5.2 cheaper |
| Cache discount | custom | none | GLM-5.2 deeper |
| Speed | ~45 tok/s | ~50 tok/s | GLM-5.2 faster |
| TTFT | 0.7s | 0.6s | GLM-5.2 snappier |
| Function calling | 80 | 80 | Tie |
| Refusal rate | ~6% | ~8% | Kimi K3 less restrictive |
| English | 75 | 70 | Kimi K3 |
| Chinese | 92 | 90 | Kimi K3 |
| Modalities | text, image | text, image | Same |
| Open weights | Yes | Yes | Both open |
| Fine-tuning | No | Yes | |
| Free tier | Kimi App free; free quota on the open platform | ChatGLM free; free quota on the open platform | |
| SOC2 / no-train | no / yes | no / yes | |
| Private deployment | Yes | Yes | Both support it |
Fields drawn from vendor public documentation and the Modelspectra 22-model dataset; speed varies with network, concurrency and prompt length. Verify current pricing before purchase.
Dimension-by-dimension analysis
Reasoning & overall intelligence
Kimi K3 leads the overall aggregate by 4 points (77 vs 73). That is a meaningful, not marginal, edge on hard, multi-step reasoning, while GLM-5.2 remains a strong generalist that is not out of its depth on routine work.
Agentic coding
This is a modest gap: Kimi K3 scores 71 against 66. On multi-file edits, SWE-style tickets and long-horizon agent loops Kimi K3 needs fewer correction turns; GLM-5.2 is still competent for scripts and assisted completion but trails as task complexity rises.
Multimodal
Kimi K3 leads multimodal 81 vs 79. Neither emits native video, so the comparison is about parsing images and documents, not generation.
Speed & latency
GLM-5.2 is faster in interactive use: ~50 tok/s with 0.6s TTFT versus ~45 tok/s with 0.7s TTFT (about 1.1× the throughput). For conversational UIs where perceived responsiveness drives retention, that edge is a real product factor even when raw reasoning is lower.
Context: window vs usable recall
Nominal windows are 256K for Kimi K3 and 128K for GLM-5.2. Effective-context scores point the same way as window size — Kimi K3 is ahead on usable recall (99 vs 88), so prefer it for long-document work where details cannot be missed.
Price & total cost
GLM-5.2 is the cheaper API at $1.4/$4.4 versus Kimi K3 at $3/$15 per 1M input/output tokens — list input is about 2.1× lower.
Chinese vs English
English: Kimi K3 75 vs GLM-5.2 70. Chinese: 92 vs 90. For Chinese-language production, Kimi K3 is the stronger pick. Note that non-Chinese models generally require overseas network access for their APIs.
Tool use & ecosystem
Function-calling score: Kimi K3 80 vs GLM-5.2 80, so Kimi K3 has the edge on structured tool use. Fine-tuning is available from GLM-5.2. Factor in existing SDK/plugin familiarity — switching cost often outweighs a few-point tool-use gap.
Cost worked example — same workload, real token math
Assume a production workload of 100M input + 30M output tokens per month, with 90% of input tokens served from cache. Figures use public list prices.
| Scenario · per month | Kimi K3 | GLM-5.2 | Gap |
|---|---|---|---|
| List priceno cache applied | $750100M in × $3 + 30M out × $15 | $272100M in × $1.4 + 30M out × $4.4 | 2.76×gap |
| With caching90% of inputs cache-hit | $750no published cache discount | $272no published cache discount | 2.76×gap |
Illustrative model; your input/output mix and cache-hit ratio change the result. Prices are list rates before any enterprise agreement.