No models match those filters. Clear the search, or reset a filter above.
}
;
};
SambaStack supports a variety of models that can be deployed to both on-prem and hosted environments. Contact your system administrator to determine which models are available on your deployment. For definitions of the **Preview** and **Production** designations, see the [Glossary](/docs/en/resources/glossary).
## Deployment options
When deploying models in SambaStack, administrators can select from various context length and batch size combinations.
* Smaller batch sizes provide higher token throughput (tokens/second).
* Larger batch sizes provide better concurrency for multiple users.
SambaStack supports two deployment configurations – high-interactivity and high-throughput. See [High-throughput deployment](/docs/en/v2.1.2/sambastack/service-administration/model-deployment/deploying-models-and-bundles/deployment-configurations) for when to use each.
## Finding models on your cluster
You can run the following command to discover available models in your cluster:
```shellscript theme={}
kubectl -n get models
```
The command lists Kubernetes resource names (`metadata.name`). Inference requests must use the serving name (`spec.name`), which is what the **Model ID** headings below use.
```yaml theme={}
apiVersion: sambanova.ai/v1alpha1
kind: Model
metadata:
name: gemma-4-31b-it # Kubernetes resource name – what kubectl get models returns
spec:
name: gemma-4-31B-it # serving name – use this in API requests
```
The two often differ only in case or punctuation, so the mismatch is easy to miss. In the example above they differ by a single character.
To read the serving name of a deployed model, run:
```shellscript theme={}
kubectl -n describe model
```
A model can also define `spec.aliases`, additional names that route to the same model. For the full `Model` field reference, see [Deploy custom checkpoints](/docs/en/v2.1.2/sambastack/service-administration/model-deployment/deploying-models-and-bundles/deploy-custom-checkpoints).
Text to speech is one case where the names diverge: address `/v1/audio/speech` as `qwen3-tts`, not by the talker or vocoder resource names. See [Text to speech](/docs/en/features/speech).
## All supported models
Search or filter to find a model, then select it to jump to its full configuration.
## Meta
### Meta-Llama-3.3-70B-Instruct
### Meta-Llama-3.1-8B-Instruct
### Meta-Llama-3.1-70B-Instruct
### Meta-Llama-3.1-405B-Instruct
### Llama-4-Maverick-17B-128E-Instruct
## MiniMax
### MiniMax-M3
### MiniMax-M2.7
### MiniMax-M2.5
## Mistral AI
### Mistral-Large-3-675B-Instruct-2512
## DeepSeek
### DeepSeek-R1-0528
### DeepSeek-R1-Distill-Llama-70B
### DeepSeek-V3-0324
### DeepSeek-V3.1
### DeepSeek-V3.2
### DeepSeek-V3.1-Terminus
## OpenAI
### gpt-oss-120b
### gpt-oss-20b
### Whisper-Large-v3
## Google
### gemma-3-27b-it
### gemma-3-12b-it
### gemma-4-31B-it
## Alibaba Cloud
### Qwen3-235B-A22B-Instruct-2507
### Qwen3-32B
### Qwen3-TTS-Talker
### Qwen3-TTS-Vocoder
## Tokyotech-llm
### Llama-3.3-Swallow-70B-Instruct-v0.4
## Other
### E5-Mistral-7B-Instruct
## Recommended model bundles
In SambaStack, a bundle is a packaged deployment that groups one or more models together with their associated configurations, such as batch size and sequence length. A single model can also be deployed on its own by pairing it with a model profile, without creating a bundle.
For example, deploying the `Meta‑Llama‑3.3‑70B` model with a batch size of 4 and a sequence length of 16K tokens constitutes a single configuration. A bundle, however, can contain multiple such configurations, either for the same model or for different models.
SambaNova’s RDU technology enables several models and configurations to be loaded simultaneously in a single deployment. This allows you to switch instantly between models and between batch‑/sequence‑size profiles as needed. In contrast to traditional GPU systems, where deployments are typically single‑model and static, SambaStack supports multi‑model, multi‑configuration bundles. This approach delivers higher efficiency, greater flexibility, and increased throughput while preserving low latency.
You can run the following command to discover available bundles in your cluster:
```shellscript theme={}
kubectl -n get modelbundles
```
The table below lists the recommended bundles for the models currently available in SambaStack. Each entry pairs a model with its recommended deployment bundle.
If the bundles listed below do not satisfy your inference requirements, you can create [custom bundles](/docs/en/v2.1.2/sambastack/service-administration/model-deployment/deploying-models-and-bundles/create-a-custom-bundle) that combine any mix of models and configurations so long as they fit in DDR memory.
### Suggested bundles per model
For each model, this section lists the suggested bundle for typical use and any alternative bundles that trade off context length, batch size, or modality support. See [Bundle configurations](#bundle-configurations) below for the seq length/batch size details of each bundle.
#### Meta
**`Meta-Llama-3.3-70B-Instruct`**
**Suggested:** `70b-3dot3-ss-4-8-16-32-64-128k`
**Alternatives:** `70b-3dot3-ss-full-whisper`, `us-agentic-rag-1-1`, `e5-mistral-70b-64k-128k`
**`Meta-Llama-3.1-8B-Instruct`**
**Suggested:** `us-agentic-rag-1-1`
**Alternatives:** `qwen3-32b-llama405b-s-m`
**`Meta-Llama-3.1-405B-Instruct`**
**Suggested:** `qwen3-32b-llama405b-s-m`
**`Llama-4-Maverick-17B-128E-Instruct`**
**Suggested:** `llama-4-medium-8-16-32-64-128k`
**Alternatives:** `llama-4-medium-ss-16k-bs24`, `us-agentic-rag-1-1`
#### MiniMax
**`MiniMax-M3`**PREVIEW
**Suggested:** `minimax-m3-32-64-128-256-512k-1m` (full 32K–1M context range)
**Alternatives:** `minimax-m3-512k` (32K–512K context range), `minimax-m3-32k`
**`MiniMax-M2.7`**
**Suggested:** `dyt-minimax-m2p7-32-64-160-192k-pc` (adds [prompt caching](/docs/en/v2.1.2/sambastack/service-administration/performance/prompt-caching))
**Alternatives:** `dyt-minimax-m2p7-32k-pc`, `dyt-minimax-m2p7-32-64-160-192k`, `dyt-minimax-m2p7-32-160-192k`, `dyt-minimax-m2p7-32k-v2`
**`MiniMax-M2.5`**
**Suggested:** `dyt-minimax-m2p5-32-160k`
**Alternatives:** `dyt-minimax-m2p5-32k`
#### Mistral AI
**`Mistral-Large-3-675B-Instruct-2512`**PREVIEW
**Suggested:** `mistral-large-3-fp8-8-16-32k`
**Alternatives:** `mistral-large-3-fp8-8k`
#### DeepSeek
**`DeepSeek-R1-0528`**
**Suggested:**
* `deepseek-4in1-fp8-128k` (higher context length)
**Alternatives:** `deepseek-r1-v3-fp8-8k`, `deepseek-r1-v31-fp8-8k`
**`DeepSeek-V3-0324`**
**Suggested:**
* `deepseek-4in1-fp8-128k` (higher context length)
**Alternatives:** `deepseek-r1-v3-fp8-8k`, `deepseek-v3-v31-fp8-8k`, `deepseek-v3-v3termi-fp8-8k`
**`DeepSeek-V3.1`**
**Suggested:**
* `deepseek-4in1-fp8-128k` (higher context length)
**Alternatives:** `deepseek-r1-v31-fp8-8k`, `deepseek-v3-v31-fp8-8k`
**`DeepSeek-V3.1-Terminus`**
**Suggested:**
* `deepseek-4in1-fp8-128k` (higher context length)
**Alternatives:** `deepseek-v3-v3termi-fp8-8k`
#### OpenAI
**`gpt-oss-120b`**
**Suggested:** `us-agentic-rag-1-1`
**Alternatives:** `cd-dyt-gpt-oss-120b-8-32-64-128k`, `gpt-gemma-whisper-mistral`
**`gpt-oss-20b`**PREVIEW
**Suggested:** `dyt-gpt-oss-20b-32-64-128k`
**`Whisper-Large-v3`**
**Suggested:** `qwen3-32b-whisper-e5-mistral`
**Alternatives:** `70b-3dot3-ss-full-whisper` (known issue: does not load within default startup time), `gpt-gemma-whisper-mistral`
#### Google
**`gemma-3-27b-it`**PREVIEW
**Suggested:** `gemma3-27b-32-128k`
**`gemma-3-12b-it`**PREVIEW
**Suggested:** `gemma3-v3`
**Alternatives:** `gpt-gemma-whisper-mistral`
**`gemma-4-31B-it`**
**Suggested:**
* `gemma-4-31b-32-128-256k` (adds 256K context; text, image, and video support)
* `gemma-4-31b-32-128k` (up to 128K context; text, image, and video support, higher throughput)
Both bundles support constrained decoding across all of their sequence lengths. Choose between them on context length and throughput, not capability.
#### Alibaba Cloud
**`Qwen3-235B-A22B-Instruct-2507`**
**Suggested:** `dyt-qwen3-235b-32-128k`
**`Qwen3-32B`**
**Suggested:** `qwen3-32b-whisper-e5-mistral`
**Alternatives:** `qwen3-32b-llama405b-s-m`
**`Qwen3-TTS-Talker`**PREVIEW
**Suggested:** `qwen3-tts-talker`
**`Qwen3-TTS-Vocoder`**PREVIEW
**Suggested:** `qwen3-tts-vocoder`
#### Other
**`E5-Mistral-7B-Instruct`**
**Suggested:** `us-agentic-rag-1-1`
**Alternatives:** `e5-mistral-70b-64k-128k`, `qwen3-32b-whisper-e5-mistral`, `gpt-gemma-whisper-mistral`
### Bundle configurations
The table below lists the configuration details for each bundle referenced above.
| Model name | Bundle name | Bundle description | Bundle configuration |
| ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :---------------------------------------------------------------------- | :-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | :------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `DeepSeek-R1-0528` / `DeepSeek-V3.1` | deepseek-r1-v31-fp8-8k |
Small to medium context length with varied batch size
[Speculative decoding](/docs/en/v2.1.2/sambastack/service-administration/performance/deploy-with-speculative-decoding) supported for `Meta-Llama-3.3-70B`
[Speculative decoding](/docs/en/v2.1.2/sambastack/service-administration/performance/deploy-with-speculative-decoding) supported for `Meta-Llama-3.3-70B`
Homogeneous bundle with constrained decoding for `gemma-4-31B-it`. Adds 32K–256K context support. Supports text, image, and video input at all sequence lengths.
Full context-length range with low to medium batch size, plus Whisper transcription
[Speculative decoding](/docs/en/v2.1.2/sambastack/service-administration/performance/deploy-with-speculative-decoding) supported for `Meta-Llama-3.3-70B-Instruct`
Homogeneous bundle containing `MiniMax-M2.7` configurations; same context range as `dyt-minimax-m2p7-32-64-160-192k-pc` without prompt caching.
| View
`MiniMax-M2.7`
Seq Length: 8K-32K, BS: 2, 4, 6, 8
Seq Length: 64K, BS: 2
Seq Length: 160K-192K, BS: 2
|
| `MiniMax-M2.7` | dyt-minimax-m2p7-32k-pc | Homogeneous bundle with prompt caching for `MiniMax-M2.7`; medium context length with high batching. | View
`MiniMax-M2.7`
Seq Length: 8K-32K, BS: 2, 4, 6, 8
|
| `MiniMax-M3`PREVIEW | minimax-m3-32k | Homogeneous bundle containing `MiniMax-M3` configurations; medium context length. Supports text, image, and video input. | View
[Speculative decoding](/docs/en/v2.1.2/sambastack/service-administration/performance/deploy-with-speculative-decoding) supported for `Meta-Llama-3.1-405B-Instruct`
|
† `cd-dyt-gpt-oss-120b-8-32-64-128k` supports sequence lengths from 8K–128K, and `dyt-gpt-oss-20b-32-64-128k` supports 32K–128K. If you require shorter context lengths (4K–16K) for either model, contact your SambaNova representative.