FreeLLMAll models › Compare
Gemma 4 31B vs kimi-k2.5-thinking
Gemma 4 31B and kimi-k2.5-thinking are about even on Arena (1443 vs 1445): in a blind vote each wins about half the time.
Gemma 4 31B1443Arena chat · #67 of 416
kimi-k2.5-thinking1445Arena chat · #62 of 416
Arena scores by category
| Category | Gemma 4 31B | kimi-k2.5-thinking | Ahead |
|---|---|---|---|
| Chat | 1443 | 1445 | about even |
| Vision | 1277 | 1269 | about even |
Scores from blind human votes on Arena. A gap under 10 points is too small to call.
Size and where to use them free
| Gemma 4 31B | kimi-k2.5-thinking | |
|---|---|---|
| Made by | moonshot | |
| Size | 31.3 B | not public |
| 4-bit download | 20 GB | — |
| Use it free | free API on OpenRouter · download and run at home · Apache 2.0 license | Modified MIT license |
Speed at home
| Hardware | Gemma 4 31B | kimi-k2.5-thinking |
|---|---|---|
| RTX 3090 | 33 tok/s | online only |
| RTX 4090 | 36 tok/s | online only |
| RTX 5090 | 62 tok/s | online only |
| RX 7900 XTX | 34 tok/s | online only |
| Mac M4 Max 128 GB | 19 tok/s | online only |
| Ryzen AI Max+ 395 128 GB | 9.4 tok/s | online only |
Estimates at 4-bit with an 8K context, 64 GB of system RAM for offload, calibrated on 44 real llama.cpp tests.
Try them yourself
Ask both the same question side by side in Try & compare, or vote blind between free models in Free Arena.
More comparisons
- Gemma 4 31B vs MiMo-V2.6-Pro
- Gemma 4 31B vs Kimi K3 (Max)
- Gemma 4 31B vs GLM-5.3 (Max)
- Gemma 4 31B vs GLM-5.3-Flash
- Gemma 4 31B vs GLM-5.2 (Max)
- Gemma 4 31B vs MiMo-V2.5-Pro
- kimi-k2.5-thinking vs MiMo-V2.6-Pro
- kimi-k2.5-thinking vs Kimi K3 (Max)
- kimi-k2.5-thinking vs GLM-5.3 (Max)
- kimi-k2.5-thinking vs GLM-5.3-Flash