> For the complete documentation index, see [llms.txt](https://docs.nerve-protocol.com/llms.txt). Markdown versions of documentation pages are available by appending `.md` to page URLs; this page is available as [Markdown](https://docs.nerve-protocol.com/overview/soma.md).

# Soma

**Soma** is Nerve Protocol's flagship language model. It is a world-class **32B** dense, all-active model with a 256K-token context window that runs entirely inside a sealed Trusted Execution Environment (TEE). Soma refuses nothing and exposes nothing: your prompts and outputs never leave the enclave, so even the host running the hardware cannot read them.

Soma is the intelligence layer behind your Personal AI. Where most frontier models are wrapped in guardrails and routed through someone else's servers, Soma is built for operators who want frontier-class quality with zero refusals and full ownership of the weights, keys, and data. It delivers massive compute power while running comfortably on a single consumer GPU.

***

### Why Soma

* **Fully uncensored, zero refusals.** No alignment guardrails, no canned denials. Soma answers the question you actually asked.
* **Enclave-native.** Every inference runs inside a sealed TEE. The host never sees your prompts, and nothing leaves the enclave unencrypted.
* **You own it.** You hold the weights and the keys. Self-host Soma or run it on the decentralized Nerve grid.
* **$NPX-gated access.** API access is metered through the $NPX token, keeping inference decentralized and operator-owned.
* **Architectural stability.** A pure dense architecture offers extreme reliability with no mixture-of-experts routing variance.

***

### Specifications

| Property       | Soma                          |
| -------------- | ----------------------------- |
| Parameters     | 32B (dense, all-active)       |
| Context window | 256K tokens                   |
| Execution      | Sealed TEE (enclave)          |
| Refusal rate   | 0%                            |
| Weights        | Open, user-owned (Apache 2.0) |
| Access         | $NPX-gated API or self-host   |

***

### Benchmark Performance

Against its closest same-class rival, Qwen 3.6, Soma holds the edge across the board—dominating on vision while securing definitive leads in graduate-level reasoning and competition math.

| Benchmark             | Soma (32B) | Qwen3.6 35B-A3B | GLM-5.1 | Llama 4 Scout |
| --------------------- | ---------- | --------------- | ------- | ------------- |
| **MMLU Pro**          | **85.9%**  | \~85.2%         | —       | —             |
| **AIME 2026**         | **93.4%**  | 92.7%           | 95.3%   | —             |
| **GPQA Diamond**      | **86.8%**  | 86.0%           | \~86.4% | 74.3%         |
| **LiveCodeBench v6**  | **82.1%**  | —               | —       | —             |
| **MMMU Pro (vision)** | **76.9%**  | 75.1%           | —       | —             |
| **Context**           | 256K       | 256K            | 200K    | 10M           |

Key wins:

* **Best-in-class vision:** Tops MMMU Pro and MATH-Vision among self-hostable models.
* **Unmatched class performance:** Edges out Qwen 3.6 in pure text reasoning and math — undisputed crown in the 30B–35B tier.
* **Elite at scale:** Beats Llama 4 Scout outright across reasoning, math, and vision under Apache 2.0.

***

### How Soma Compares

Soma is built to match leading open-weight models such as Google's Gemma 4 on quality and scale, while adding the things closed and centrally hosted models cannot offer: zero censorship, enclave-level privacy, and true ownership.

|                                  | Soma | Gemma 4 26B | Llama 3.1 70B | Dolphin 2.9 | Gemma 2 27B |
| -------------------------------- | ---- | ----------- | ------------- | ----------- | ----------- |
| Fully uncensored, zero refusals  | ✅    | ❌           | ❌             | ◐           | ❌           |
| Private enclave (TEE) execution  | ✅    | ❌           | ❌             | ❌           | ❌           |
| Prompts never leave your control | ✅    | ❌           | ❌             | ❌           | ❌           |
| You own the weights and keys     | ✅    | ✅           | ◐             | ◐           | ◐           |
| Decentralized inference grid     | ✅    | ❌           | ❌             | ❌           | ❌           |
| $NPX-gated API access            | ✅    | ❌           | ❌             | ❌           | ❌           |
| Parameters                       | 32B  | 26B         | 70B           | 8B          | 27B         |
| Context window                   | 256K | 256K        | 128K          | 8K          | 8K          |
| Open weights                     | ✅    | ✅           | ✅             | ✅           | ✅           |

*✅ full · ◐ partial · ❌ none. Comparison reflects default configurations of publicly available open-weight models; capabilities vary by deployment.*

***

{% hint style="info" %}
*Note:* Soma runs inside the same hardware-enforced enclave as the rest of Nerve Protocol. Your prompts, context, and outputs stay sealed; only you can read them.
{% endhint %}


---

# Agent Instructions
This documentation is published with GitBook. GitBook is the documentation platform designed so that both humans and AI agents can read, navigate, and reason over technical content effectively. Learn more at gitbook.com.

## Querying This Documentation
If you need additional information that is not directly available in this page, you can query the documentation dynamically by asking a question.

Perform an HTTP GET request on the current page URL with the `ask` query parameter, and the optional `goal` query parameter:

```
GET https://docs.nerve-protocol.com/overview/soma.md?ask=<question>&goal=<endgoal>
```

`ask` is the immediate question: it should be specific, self-contained, and written in natural language.
`goal` is optional and describes the broader end goal you are ultimately trying to accomplish on behalf of the user. GitBook uses it to tailor the answer towards what is most useful for that goal.

The response will contain a direct answer to the question and relevant excerpts and sources from the documentation.

Use this mechanism when the answer is not explicitly present in the current page, you need clarification or additional context, or you want to retrieve related documentation sections.