# Ollama Cloud Coding Models — Comparison Notes

Research snapshot: June 2026. Built from the Ollama cloud catalog, official model cards, technical reports, and third-party comparisons.

## Candidate models

| Model | Type | Active params | Context | License | Best for |
|---|---|---|---|---|---|
| glm-5.2:cloud | API / MoE | large | 1M | MIT | Long-horizon agentic engineering, repo-wide work |
| kimi-k2.7-code:cloud | Coding-specialized | large | 256K | proprietary via Moonshot | MCP-heavy coding agents, vision+code |
| qwen3.6:27b | Dense local/cloud | 27B | ~128K | open | Best dense coding model you can also run locally |
| qwen3.6:35b-a3b | MoE local/cloud | 35B total / 3B active | ~128K | open | Efficient alternative to 27B dense |
| qwen3-coder-next:cloud | MoE | 80B total / ~3B active | long | Apache 2.0 | High-volume, cost-sensitive coding tasks |
| qwen3-coder:480b-cloud | MoE | 480B total / small active | long | open | Highest-capacity Qwen coder |
| nemotron-3-ultra:cloud | MoE | 550B total / 55B active | long | open | Long-running multi-agent orchestration |
| gemma4:31b-cloud | Dense | 31B | large | open | Strong all-rounder, vision/audio/code |
| minimax-m2.7:cloud | MoE | M2-series | — | licensed | Efficient agentic coding and office tasks |

## Benchmark snapshot

| Model | SWE-Bench Verified | SWE-Bench Pro | Terminal-Bench 2.0 | LiveCodeBench v6 | Notes |
|---|---|---|---|---|---|
| GLM-5.2 | — | 62.1 | 81.0 | — | Strongest public scores on this list |
| Kimi K2.7-Code | — | — | — | — | ~30% lower thinking tokens than K2.6; MCP Mark Verified 81.1 |
| Qwen3.6-27B | 77.2 | 53.5 | 59.3 | — | Surpasses Qwen3.5-397B-A17B on coding |
| Qwen3-Coder-Next | ~70.6 | — | — | — | 80B/3B MoE, very efficient |
| Nemotron 3 Ultra | 65–70.4 | — | 54.0 | — | Consistent across agent frameworks |
| Gemma 4 31B | — | — | — | 80.0 | Codeforces ELO 2150 |
| MiniMax M2.7 | — | 56.22 | 57.0 | — | Matches GPT-5.3-Codex per maker |

## Pricing / usage notes

- Ollama Cloud bills by GPU time, not per token. Larger / MoE models consume plan allowance faster.
- Free plan is light usage; Pro ($20/mo) is day-to-day work; Max ($100/mo) is heavy sustained agent usage.
- Some models only have `:cloud` tags; verify tag availability in the Ollama catalog before recommending.

## Recommendation heuristics

- Raw benchmark leader with huge context: GLM-5.2
- Coding-specialized efficiency with vision: Kimi K2.7-Code
- Dense model you can also run locally: Qwen3.6-27B
- Cheapest high-volume coding: Qwen3-Coder-Next
- Long-running agent orchestration throughput: Nemotron 3 Ultra
- Well-rounded multimodal coding: Gemma 4 31B

## Sources

- https://ollama.com/search?c=cloud
- https://ollama.com/library/glm-5.2
- https://ollama.com/library/kimi-k2.7-code
- https://ollama.com/library/qwen3.6
- https://ollama.com/library/qwen3-coder
- https://ollama.com/library/qwen3-coder-next
- https://ollama.com/library/nemotron-3-ultra
- https://ollama.com/library/gemma4
- https://ollama.com/library/minimax-m2.7
- https://qwen.ai/blog?id=qwen3.6-27b
- https://developer.nvidia.com/blog/nvidia-nemotron-3-ultra-powers-faster-more-efficient-reasoning-for-long-running-agents/
- https://www.llmreference.com/compare/glm-5.2/kimi-k2-7-code