GLM 5.3 vs Kimi K3
A side-by-side of live API prices, providers, speed, usage and benchmarks. No verdict: the right choice depends on your workload.
GLM 5.3 costs $0.50 per 1M input and $1.51 per 1M output tokens at the standard tier, via Sail Research (fp8); Kimi K3 costs $0.504 and $8.10, via Sail Research (fp4). At a 3:1 input:output mix, GLM 5.3 is 69% cheaper than Kimi K3.
Why these two are compared
Two large models from Chinese labs served by many independent providers: Z.ai’s GLM 5.3 for complex software engineering and long-horizon agent tasks, Moonshot AI’s Kimi K3, an open-weight multimodal reasoning model.
Side by side
| Measure | GLM 5.3 | Kimi K3 |
|---|---|---|
| Developer | Z.ai (Zhipu) | Moonshot AI |
| Released | 2026-08-18 | 2026-07-16 |
| Standard price, input / output per 1M | $0.50 / $1.51 | $0.504 / $8.10 |
| Blended price (3:1) | $0.7513 | $2.40 |
| Cheapest standard provider | Sail Researchfp8 | Sail Researchfp4 |
| Cache read per 1M | $0.119 | $0.18 |
| Providers at the standard tier | 36 | 18 |
| Context window (tokens) | 1,048,576 | 1,048,576 |
| Max output (tokens) | 943,718 | 943,718 |
| Input modalities | text | text, image, video |
| Long-prompt pricing | No threshold listed | No threshold listed |
| Fastest endpoint | 183 tok/sFireworks AI (Fast tier) | 153 tok/sMakora (fp4) |
| Lowest latency | 239 msModelRun (fp4) | 658 msInferenceNet (Fast tier, fp4) |
| OpenRouter tokens (2026-10-16) | 942.62B | 203.41B |
| AA intelligence | 44.8 | 43.6 |
| AA coding | 74.8 | 76.2 |
| AA agentic | 53.1 | 50.0 |
“Not reported” means the source publishes no value. AA scores are the Artificial Analysis indexes reported through OpenRouter (CC BY 4.0), not SF·SIX measurements.
Cost for three example workloads
| Workload | GLM 5.3 / month | Kimi K3 / month |
|---|---|---|
| Support chatbot50,000 requests × 1,500 in / 400 out tokens | $67.60 | $199.80 |
| RAG / research assistant20,000 requests × 8,000 in / 700 out tokens | $101.07 | $194.04 |
| Batch summarisation10,000 requests × 10,000 in / 500 out tokens | $57.53 | $90.90 |
Standard headline prices; caching and long-context surcharges not included. Price your own workload.
GLM 5.3 providers
| Provider and terms | Input / 1M | Output / 1M | Throughput | Latency |
|---|---|---|---|---|
| Sail Researchfp8 | $0.50 | $1.51 | 81 tok/s | 1.31 s |
| Inceptronfp4 | $0.55 | $1.51 | 97 tok/s | 531 ms |
| DekaLLM | $0.18 | $2.95 | 80 tok/s | 1.28 s |
| Baidufp8 | $0.5992 | $1.88 | 68 tok/s | 784 ms |
| DeepInfrafp4 | $0.5625 | $2.50 | 70 tok/s | 1.58 s |
| Tenstorrentfp8 | $0.60 | $2.50 | 86 tok/s | 475 ms |
| Novita AIfp8 | $0.70 | $2.20 | 45 tok/s | 1.72 s |
| SiliconFlowfp8 | $0.70 | $2.20 | 61 tok/s | 2.00 s |
| IO.NETfp8 | $0.49 | $2.90 | 60 tok/s | 1.84 s |
| Relace | $0.0632 | $4.40 | 86 tok/s | 1.79 s |
Kimi K3 providers
| Provider and terms | Input / 1M | Output / 1M | Throughput | Latency |
|---|---|---|---|---|
| Sail Researchfp4 | $0.504 | $8.10 | 74 tok/s | 1.74 s |
| InferenceNetfp4 | $0.304 | $11.94 | 83 tok/s | 879 ms |
| Relacefp4 | $0.27 | $14.00 | 45 tok/s | 2.10 s |
| Chutesmxfp4 | $1.85 | $9.50 | 11 tok/s | 5.05 s |
| Morphfp8 | $0.80 | $14.90 | 84 tok/s | 1.49 s |
| Makorafp4 | $1.53 | $12.75 | 153 tok/s | 812 ms |
| AkashMLfp4 | $1.20 | $14.00 | 58 tok/s | 708 ms |
| Phala | $2.55 | $12.75 | 74 tok/s | 1.65 s |
| DigitalOcean | $2.55 | $12.95 | 44 tok/s | 1.31 s |
| DeepInframxfp4 | $2.85 | $14.25 | 58 tok/s | 937 ms |
Sources and method
Prices are USD per 1M tokens at the standard service tier in the default region: the cheapest qualifying provider per model, with flex, batch, priority and fast endpoints and unusually low outliers never used as the headline. “Blended” weights input and output 3:1. Prices: OpenRouter, observed at the times on each model page. Speed: OpenRouter, previous 30 minutes. Usage: OpenRouter rankings, latest full UTC day (CC BY 4.0).