Choose theme
AI Resources
Ornith 1.5
Ornith 1.5 is a reasoning-model family for coding and agent work, with 9B dense, 35B-A3B and 397B mixture-of-experts checkpoints.
The publisher describes training that generates tasks, task-specific scaffolds and solution rollouts. The collection also offers quantized formats for different serving setups; an agent still supplies the surrounding tools and workflow. Use this as a first read, not a recommendation. Open the original project before trusting details like terms, limits, privacy, cost, setup, or safety.
What it is
Supply the model beneath a coding agent
A builder can serve an Ornith checkpoint through an OpenAI-compatible endpoint and connect a coding client. The agent layer still decides how to use files, commands and returned tool calls.
Why it stands out
Task generation is part of training
The described training loop proposes new tasks as well as scaffolds and solutions, instead of only learning from a fixed task set. That describes how the checkpoint was trained, not a promise that your deployed model keeps learning from your work.
Availability
Several scales and serving formats
The official collection includes the three model scales and GGUF, FP8, NVFP4 and MLX variants. Use the card for the exact checkpoint and format; a recipe for one runtime does not apply unchanged to all of them.
Why it matters
What makes it useful
If you want to change the model behind a coding workflow, Ornith supplies checkpoints and serving examples without requiring you to replace the whole agent. Whether that pairing completes your own repository tasks still depends on the chosen client, runtime and checks.
What to know
Where it fits
The model produces reasoning, answers and tool-call output. A server parses that output, and an agent executes the requested tools. A checkpoint download by itself does not provide a working coding assistant.
Notable points
What stands out
The 35B-A3B card uses qwen3_xml as the vLLM tool-call parser and qwen3_coder for SGLang, with a separate qwen3 reasoning parser. The server's parsed tool_calls field is what the client consumes, so copying only the model name misses part of the agent connection.
Before using
What to review
Budget memory for the full chosen checkpoint and context, not only its active-parameter count. Quantization and offloading change the deployment.
Match the serving recipe, parsers and sampling settings to the exact model and runtime.
The 35B-A3B card notes that static YaRN scaling applies to every request and can slightly hurt ordinary-length output quality. Choose the longer-window setting for tasks that need it.
Benchmark numbers are publisher-reported, with different harnesses and settings; the announcement says results are averaged over five runs. They do not predict your own agent's success.
Keep repository changes, tool permissions and test results under your normal review; the model card is not a guarantee of correct code.
Reader fit
Who may find it relevant
Builders serving their own coding model or testing a different checkpoint behind an existing agent.
Readers choosing between smaller dense models, sparse models and larger server setups.
Someone wanting a ready-to-use coding app will still need a runtime and agent interface.
Editorial note
Why LifeHubber lists it
For a coding session that needs more material in one request, the 35B-A3B card includes a longer-context serving recipe alongside its ordinary setup. That gives you a concrete way to try a larger working window behind an existing coding client. Choose the window for the task; a short edit need not use the same context configuration as a long repository excerpt.
Source links
Source materials
Reader note
Before relying on this entry
LifeHubber lists entries to help readers inspect AI projects, not to endorse them or prove they are safe, suitable, accurate, maintained, or right for a specific use. We do not verify every entry in depth. Before relying on anything listed, review the original materials, terms, privacy practices, limits, and risks that matter for your situation.
What to explore next
Separate the model, the agent loop, and the test.
A coding-agent result depends on more than the checkpoint. Continue by comparing model routes, seeing what the agent layer adds, and checking the evidence behind agent rankings.
More in AI Models
Keep browsing this category
Explore more AI model resources.
Gemma 4
google/gemma-4
A Google DeepMind Gemma 4 model family collection with public checkpoints including Gemma 4 12B, a dense multimodal model Google describes around local agentic workflows, native audio input, and encoder-free vision/audio handling.
MiniMax H3 Integrations
MiniMax-AI/awesome-minimax-h3-integration
A community-maintained MiniMax H3 integration index that maps checkpoints, hardware and VRAM starting points, runtimes, ComfyUI nodes, prompting tools, acceleration routes, and deployment options.
OLMoEarth
allenai/olmoearth
An Ai2 remote-sensing foundation model family for satellite imagery and planetary-scale mapping, with v1.2 Nano, Tiny, Small, and Base models, model weights, training code, and a technical report.
For project maintainers
Listed here? You can use the badge.
If you maintain a project with a current LifeHubber listing, you may add the optional “Listed on LifeHubber AI Resources” badge to its README, docs, or website. No introduction or permission request is needed.
