Model Support Roadmap

Model Support Roadmap

Last updated: 2026-08-04.

This page mirrors the repository roadmap. The authoritative status and delivery sequence source is docs/MODEL-SUPPORT.json. Current source code takes precedence if any Wiki page disagrees.

Delivery Sequence

Order	Workstream	Project status	Entry gate
1	XAIOS	In Progress	Finish the core OS, portable engine, model-v2 integration, platform services, hardware readiness, and release gates.
2	Qwen 3.6 27B Support	Blocked	Starts only after the XAIOS completion gate.
Later	Kimi K3 Support	Backlog	Queued behind XAIOS and Qwen unless explicitly reprioritized.
Later	DeepSeek V4 Flash 0731 Support	Blocked	Also blocked on authoritative release and source verification.
Later	GLM 5.2 Support	Backlog	Queued behind XAIOS and Qwen unless explicitly reprioritized.

Support Status

Model or path	Status	Current boundary
Deterministic QEMU model-v1 path	Fixture only	Validates OS/runtime contracts. It is not transformer inference or a hardware benchmark.
xaios.model.v2 tooling	Interface only	Package parser/writer and interface tests exist; no production model executes from it.
Qwen 3.6 27B	Interface only	Next model target after XAIOS; tokenizer, logits and deterministic decode parity remain incomplete.
Kimi K3 text	Interface only	Queued behind Qwen; KDA, Gated MLA, exact top-16 MoE, shared experts and MXFP4 remain unimplemented.
Kimi K3 multimodal	Roadmap only	Vision preprocessing, tower, projection and multimodal parity are a separate later gate.
DeepSeek V4 Flash 0731	Roadmap only	Exact official release, configuration and tokenizer sources must be verified first.
GLM 5.2	Roadmap only	Importer, tokenizer, operators, state and parity work have not started.

Fixture only, Interface only, and Roadmap only are not support claims. A model becomes supported only after official tokenizer, tensor import, operator/state, logits, deterministic decode, session continuity, and physical-hardware gates pass.

Workstream Boundaries

XAIOS

Complete the production-width memory/NUMA/storage paths, typed model state, full x86_64 service parity, release/security gates, and physical-hardware entry evidence before model implementation begins. The portable service boundary, no-copy model mappings, direct range I/O, ARM worker and FP/SIMD context gates, and x86 AP/ring-3/XSAVE/modern-VirtIO foundations are implemented at their documented QEMU/hosted boundary. See docs/PLATFORM-SUPPORT.json.

Qwen 3.6 27B

Qwen is workstream 2 and the first real-model target. It remains blocked until XAIOS completes. The implementation then proceeds through a smaller compatible Qwen-family checkpoint before 27B tokenizer/logits/decode and hardware gates.

Later Model Workstreams

Kimi K3, DeepSeek V4 Flash 0731, and GLM 5.2 remain queued behind XAIOS and Qwen unless explicitly reprioritized. Kimi text and multimodal acceptance are separate. DeepSeek also remains blocked on authoritative source identity.

Detailed Plans

• Repository implementation roadmap
• Model-v2 specification
• Benchmark contract
• GitHub Project