Skip to content

Model catalog & specs

Waired bundles a catalog of models spanning laptops to multi-GPU servers. On first run it auto-selects one that fits your hardware. You can switch at any time — see Switch the bundled model.

The catalog is split two ways so you can jump straight to the rows that apply to your machine:

  1. By runtime — Ollama vs vLLM. Your OS and GPU decide which engine serves the model, so each runtime has its own section. Read only yours:

    • Ollama — macOS (Metal), Windows, Linux, CPU, and low-VRAM / integrated GPUs. Ships GGUF (quantized) builds; the number to watch is system RAM.
    • vLLM — NVIDIA (CUDA) and AMD (ROCm) discrete-GPU servers on Linux. Ships safetensors (awq/fp8/mxfp4) builds; the number to watch is GPU VRAM.

    A family that ships builds for both engines appears in both sections.

  2. By architecture — Dense vs MoE. Within each runtime, Dense models come first, then Mixture-of-Experts (MoE):

    • Dense — every parameter runs on every token. Best where you have compute / VRAM headroom (a discrete GPU), and smaller on disk for a given quality.
    • MoE — huge total size but only a few billion parameters are active per token, so decode stays fast. Best on memory-rich unified-memory machines (Apple Silicon, AMD Strix Halo) that can hold a large model in RAM.

Quality tier is a 1–100 ranking (higher is better); the Auto-Selector serves the highest tier that fits your machine and can hold a large coding context (~200k tokens) — coding agents routinely fill 100k+ tokens of context, so a model that would truncate it loses to a slightly smaller one with the full window. On hardware where no model can serve that window, Waired picks the best fit anyway and says so in waired status. Whatever wins, clients can address it by the stable handle waired/default (see Chat clients).

The tables below are generated at build time from each model’s bundled manifest (min_ram_gb / min_vram_mb / quality_tier / param_count / active_params), so they always match the catalog the client ships.

For Mac, Windows, CPU, and low-VRAM / integrated GPUs. The memory column is the minimum system RAM for the GGUF (q4) build; smaller machines are listed first.

Every parameter runs each token — pick by RAM, then quality.

ModelParamsQuality tierOllama RAM
qwen2.5-coder-0.5b-instruct494M102 GB
qwen3.5-0.8b800M122 GB
qwen2.5-coder-3b-instruct3.1B314 GB
qwen3.5-2b2B274 GB
qwen2.5-coder-7b-instruct7.6B508 GB
qwen3.5-4b4B428 GB
qwen3.5-9b9B5212 GB
qwen2.5-coder-14b-instruct14.7B5816 GB
qwen3.5-27b27B6924 GB
qwen3.6-27b27B7224 GB

Large on disk but fast to decode — ideal for big unified-memory machines (Apple Silicon, Strix Halo).

ModelParams (active)Quality tierOllama RAM
gpt-oss-20b20.9B (3.6B active)6216 GB
qwen3-coder-30b-a3b-instruct30.5B (3.3B active)6832 GB
qwen3.5-35b-a3b35B (3.3B active)7332 GB
qwen3.6-35b-a3b35B (3.3B active)9032 GB
qwen3-coder-next-80b-a3b-instruct80.1B (3.3B active)8656 GB
gpt-oss-120b117B (5.1B active)8896 GB
qwen3.5-122b-a10b122B (10B active)83128 GB
qwen3-coder-480b-a35b-instruct480B (35B active)95320 GB

For NVIDIA (CUDA) and AMD (ROCm) discrete-GPU servers on Linux. The memory column is the minimum GPU VRAM for the quantized (awq/fp8/mxfp4) build; smaller GPUs are listed first. On hosts with several identical NVIDIA GPUs, vLLM shards the model across them (tensor parallelism), so their combined VRAM counts toward the requirement. On recent NVIDIA GPUs (Ada/Hopper and newer — e.g. L4, RTX 40-series), Waired serves the KV cache in fp8, which roughly doubles the context window that fits in the same VRAM at near-identical quality. If a model’s full context window still doesn’t fit your GPU memory, Waired serves it with a reduced window and notes that in waired status / waired doctor instead of failing to start.

Compute-bound; the natural fit when you have a discrete GPU with VRAM headroom.

ModelParamsQuality tiervLLM VRAM
qwen2.5-coder-3b-instruct3.1B314 GB
qwen2.5-coder-7b-instruct7.6B508 GB
qwen2.5-coder-14b-instruct14.7B5816 GB
qwen3.6-27b27B7224 GB

Big total size with a small active footprint — pairs well with multi-GPU servers.

ModelParams (active)Quality tiervLLM VRAM
gpt-oss-20b20.9B (3.6B active)6220 GB
qwen3-coder-30b-a3b-instruct30.5B (3.3B active)6824 GB
qwen3-coder-next-80b-a3b-instruct80.1B (3.3B active)8655 GB
gpt-oss-120b117B (5.1B active)8879 GB
glm-4.5-air-106b-a12b106B (12B active)75118 GB
deepseek-v4-flash284B (13B active)93192 GB
glm-5.2744B (40B active)97547 GB
qwen3-coder-480b-a35b-instruct480B (35B active)95547 GB

The Auto-Selector chooses the highest-quality model that fits your machine’s memory. On shared-memory systems (Apple Silicon, AMD Strix Halo), MoE models with a small active parameter count are favored because decode speed tracks active params, not total size. To preview a routing decision without changing anything:

Terminal window
waired infer --explain "say hi"

Run waired models ls --detail to see this same fit + recommended-spec view for your hardware, waired models ls for disk usage, and waired runtimes status for the VRAM a loaded model is actually using.

If you run an integrated GPU (Strix Halo, Intel) and a model won’t load on Ollama, see Troubleshooting → integrated GPUs.