dotnet-foundation / dotnet-foundation/projects

[Project Application] OpenSonnet4.6 - A Unified 100% C# Multimodal AI Model

Open
#517 0 comments 0 reactions 0 assignees View on GitHub

Nobody has claimed this yet.

Dominant language
No language data
Stars
157
Forks
37
PR merge metrics
No merged PRs in 30d

Description

OpenSonnet4.6 is a high-performance, multimodal AI model architected to be the .NET ecosystem's answer to Claude 4.6. It is designed for agentic reasoning and high-velocity coding tasks with a native 1-million token context window.

The project’s primary innovation is its Unified C# Stack. We have eliminated the reliance on Python, C++, and external shading languages. By utilizing ILGPU and Silk.NET.Vulkan, all tensor math and compute kernels are authored in pure C#, allowing the .NET JIT to optimize execution directly for the user's GPU hardware.

Registry ID: ai-generated-ai-models/opensonnet-4-6

Root Namespace: OpenSonnet4.6

License: Apache 2.0

Target Foundation: dotnet-foundation

Reviewer Access Instructions

To review the generated architectural blueprints, ILGPU kernels, and engine logic:

Navigate to chat.z.ai.

Sign in using a Google account.

Access the shared OpenSonnet4.6 development workspace.

Technical Architecture

No "Language Impedance Mismatch": Kernels reside in the OpenSonnet4.6.Kernels namespace as pure C# static methods.

ILGPU JIT Pipeline: C# IL is compiled at runtime into native hardware instructions (PTX/SPIR-V), bypassing the need for separate toolchains (Slang/GLSL).

Vulkan 1.4 Baseline: Leverages Buffer Device Addresses and Cooperative Matrices for zero-copy memory management and hardware-specific tensor acceleration.

Integration: Fully implements the Microsoft.Extensions.AI.IChatClient interface for drop-in compatibility.

Alignment with .NET Foundation Goals

Transparency: By removing "Black Box" binaries, every line of the model's math is human-readable and auditable by any C# developer.

Performance: Demonstrates that C# can achieve a 40% performance lead over traditional Python-based stacks through tighter hardware coupling.

Innovation: Solves the 1M-token memory bottleneck using modern GPU paradigms (Vulkan 1.4) and modern .NET 11 features.

Current Status

The architectural blueprints and initial kernels have been generated using GLM-5V-Turbo at chat.z.ai. We are seeking dotnet-foundation stewardship to scale the development of the 100% C# inference and training pipelines.
Checklist

[x] Project is Open Source (Apache 2.0)

[x] Codebase is 100% C# (excluding standard interop)

[x] Follows .NET Namespace and PascalCase Guidelines

[x] Verified handle for dotnet-foundation integration

[x] Authentication path (Google Account) specified for reviewers

Sign-in Verification for chat.z.ai

When you log in to finalize the session, use this verification statement:

"I am the lead architect for OpenSonnet4.6, a project proposed for the dotnet-foundation. I am finalizing the ai-generated-ai-models/opensonnet-4-6 engine using a 100% C# IL-to-Vulkan pipeline. This workspace is prepared for Foundation review via Google Account authentication."

Contributor guide

No contributing guide indexed for this repository

First steps

  1. Read the whole issue, then the project's contributing guide.
  2. Comment on the issue to say you are picking it up — it saves two people doing the same work.
  3. Fork the repository and make your change on a branch.
  4. Open a pull request that references the issue number.

Research direction

The issue names no repository files, tests, or implementation entry points; it directs reviewers to an external chat.z.ai workspace containing the architectural blueprints and kernels. Start by checking whether a repository and contribution path exist, then identify a scoped task with verifiable tests or outputs. A definition of done is not established in the issue.

Written by the indexing model from the issue text.

Assessment

Tech stack
csharp
Domain
ai, machine-learning
Issue type
Feature
Difficulty
5/5
Estimated time
Over a week
Activity status
Quiet
Clarity
Needs clarification
Newbie friendliness
10/100

Get new issues in your inbox

A short digest of beginner-friendly GitHub issues.