Which open models should I use?
Actually, there’s no single right answer! Here’s a curated list based on Novita internal testing, community feedback, and external benchmarks. We recommend using it as a starting point, and we will update it regularly as new models emerge.- Model sizes are marked as Small, Medium, or Large. For MoE models, the activated parameter size is what matters for latency.
- For best latency, use small or medium models. For best quality, use large models or fine-tune medium or small models.
- Model names below are exact model IDs — copy them directly into your API calls. Each links to its model page for live pricing and a playground.
- You can explore all models in the Novita Model Library.
Code generation & reasoning
| Model ID | Size / Notes | Context Window |
|---|---|---|
moonshotai/kimi-k2.7-code | Large | 262K |
zai-org/glm-5.2 | Large | 1M |
deepseek/deepseek-v4-pro | Large | 1M |
General reasoning & planning
| Model ID | Size / Notes | Context Window |
|---|---|---|
deepseek/deepseek-v4-pro | Large | 1M |
deepseek/deepseek-v3.2 | Large | 160K |
qwen/qwen3.5-397b-a17b | Large, 17B active | 262K |
Agents, function calling & tool use
| Model ID | Size / Notes | Context Window |
|---|---|---|
zai-org/glm-5.2 | Large | 1M |
moonshotai/kimi-k2.7-code | Large | 262K |
minimax/minimax-m3 | Large | 1M |
Long context & summarization
| Model ID | Size / Notes | Context Window |
|---|---|---|
zai-org/glm-5.2 | Large | 1M |
deepseek/deepseek-v4-pro | Large | 1M |
deepseek/deepseek-v4-flash | Large, cost-effective | 1M |
Vision & document understanding
| Model ID | Size / Notes | Context Window |
|---|---|---|
moonshotai/kimi-k2.5 | Large | 262K |
qwen/qwen3.5-397b-a17b | Large, 17B active | 262K |
google/gemma-4-31b-it | Medium | 262K |
Low-latency NLU & extraction
| Model ID | Size / Notes | Context Window |
|---|---|---|
inclusionai/ling-2.6-flash | MoE, small activation | 262K |
meta-llama/llama-3.1-8b-instruct | Small | 128K |
google/gemma-4-26b-a4b-it | Medium, 4B active | 262K |
Migrating from older models
If you are still calling one of the models below, switch to its recommended replacement:| If you use | Switch to |
|---|---|
deepseek/deepseek-r1-0528 | deepseek/deepseek-v4-pro |
zai-org/glm-4.5 | zai-org/glm-5.2 |
moonshotai/kimi-k2-instruct | moonshotai/kimi-k2.7-code |
qwen/qwen3-coder-480b-a35b-instruct | moonshotai/kimi-k2.7-code |
meta-llama/llama-4-maverick / llama-4-scout, qwen/qwen2.5-vl-72b-instruct | moonshotai/kimi-k2.5, qwen/qwen3.5-397b-a17b, or google/gemma-4-31b-it |