Ollama Cloud · models
Reference · every cloud model, one table
Everything on Ollama Cloud, priced.
All 21 cloud models Ollama serves, in 22 rows1 (gpt-oss ships two sizes), with the developer behind each one, the context and parameter count you actually get, and what it is for. Click any column header to sort — click again to reverse.
Prices checked at ollama.com/pricing 24-Oct-2026 - 9:37am Pacific (16:37 UTC). Before plan credits. Page loads sorted by model, A to Z.
| Model▲ | Input2▲ | Cached▲ | Output▲ | Parent company▲ | Country3▲ | Context▲ | Params4▲ | What it is▲ |
|---|---|---|---|---|---|---|---|---|
| beam | free | — | free | Reflection | USA | 1024K | 501B | Token-efficient open model for agentic tool use, coding and reasoning. Free on Cloud. |
| clef | $0.24 | — | — | Cloudflare | USA | 64K | 27B | Decision model: takes a state plus a schema of typed questions and returns structured answers. |
| deepseek-v4-pro | $1.32half off-peak2 | $0.044 | $3.96 | DeepSeek | China | 1024K | 1,650B | Frontier MoE with three reasoning modes. |
| deepseek-v4.1-flash | $0.30half off-peak2 | $0.006 | $1.20 | DeepSeek | China | 1024K | 763B | Fast MoE for high-throughput agent work. |
| gemma4 | $0.14 | $0.05 | $0.40 | USA | 256K | 31B | Open Gemma family; the Cloud version is the 31B model, text and image. | |
| glm-5.2 | $1.40 | $0.26 | $4.40 | Z.ai | China | 976K | 756B | Previous-generation Z.ai flagship, still priced at the top of the range. |
| glm-5.3 | $1.40 | $0.26 | $4.40 | Z.ai | China | 1024K | 753B | Z.ai flagship: strongest open-weights coder, built for long-horizon agent tasks. |
| glm-5.3-flash | $0.15 | $0.03 | $0.50 | Z.ai | China | 1024K | 321B | Multimodal flash with ~18B active; near-frontier coding at a low price. |
| gpt-oss:120b | $0.15 | $0.014 | $0.60 | OpenAI | USA | 128K | 120B | OpenAI's larger open-weight MoE; stronger reasoning at the same 128K context. |
| gpt-oss:20b | $0.07 | $0.035 | $0.30 | OpenAI | USA | 128K | 20B | OpenAI's small open-weight reasoning model; only beam and the Nemotron models are cheaper in the US column. |
| kimi-k2.6 | $0.95 | $0.16 | $4.00 | Moonshot AI | China | 256K | 1,040B | Multimodal agentic model for long-horizon coding and task orchestration. |
| kimi-k2.7-code | $0.95 | $0.19 | $4.00 | Moonshot AI | China | 256K | 1,040B | Coding-focused Kimi built on K2.6; about 30% fewer thinking tokens. |
| kimi-k3 | $3.00 | $0.30 | $15.00 | Moonshot AI | China | 1024K | 2,810B | Moonshot's flagship multimodal agentic model; the most expensive row here. |
| minimax-m2.7 | $0.30 | $0.06 | $1.20 | MiniMax | China | 200K | 229B | M2-series workhorse for coding, agentic workflows and office productivity. |
| minimax-m3 | $0.60 | $0.12 | $2.40 | MiniMax | China | 512K | 428B | Coding and agentic frontier model with sparse attention and native multimodality. |
| minimax-m3.1-flash | $0.15 | $0.015 | $0.60 | MiniMax | China | 976K | 428B | Flash tier for agentic, coding and computer-use work; ~3x faster than M3. |
| mistral-large-3 | $0.50 | — | $1.50 | Mistral AI | France | 256K | 675B | Production-grade multimodal MoE; no cached-input rate published. |
| mistral-large-4 | $0.68 | $0.07 | $2.09 | Mistral AI | France | 1024K | 1,050B | Open-weight generalist MoE: 52B active of 1.05T, with a vision encoder. |
| muse-spark-open | $1.25 | $0.15 | $4.25 | Meta | USA | 1024K | 816B | Meta Superintelligence Labs' open MoE, 42B active, built for agentic coding. |
| nemotron-3-nano | $0.06 | — | $0.24 | NVIDIA | USA | 1024K | 30B | Efficient agentic model; the Cloud version is 30B with a 1M context. |
| nemotron-3-super | $0.015 | $0.015 | $0.60 | NVIDIA | USA | 256K | 120B | 120B MoE activating 12B; tuned for multi-agent throughput. |
| nemotron-3-ultra | $0.10 | $0.10 | $3.00 | NVIDIA | USA | 256K | 550B | High-throughput reasoning model for long-running agent workflows. |
Footnotes
- Why 22 rows for 21 models. Ollama lists 21 cloud models; gpt-oss is one entry there but ships two sizes with different prices, so it gets two rows here. beam is free on Cloud, shown at $0. ↑
- Off-peak halves two rows. deepseek-v4.1-flash and deepseek-v4-pro bill at half the listed input, cached and output rates outside 12:00–18:00 UTC on weekdays, and all weekend. All other rows are flat. ↑
- Country is developer's HQ. Ollama hosts models primarily in the US but states they may route to Europe for additional capacity. Parent company country is named here. ↑
- Parameter counts. Totals, not active — mixture-of-experts models activate far fewer per token. From each model's Ollama page or its model card. Where a family ships several sizes, the row shows the size of the Cloud version — gemma4 is 31B, nemotron-3-nano is 30B. muse-spark-open is 816B on Meta's model card and repo name (Muse-Spark-Open-816B-A42B); Ollama's own size panel reads 822B, and the row keeps the model card's 816B. ↑