Released 2026-07-16 · general agent
Deployment evidence
Ant Group InclusionAI · 100B non-embedding; 103B repository metadata total · sparse MoE; exact active count not published active · 128K tokens
Best-fit scenarios
Diffusion-language-model research checkpoint for agent reasoning, tool use and iterative editing, with an executable Transformers path but no vendor-confirmed production server yet.
Not recommended
- Advertising SGLang as vendor-supported while the card says support is coming soon
- Treating the research generation loop as a production API topology
Diffusion agent research
Vendor-statedThe vendor presents LLaDA2.2-flash as a diffusion-language-model checkpoint for reasoning, coding and agent tasks.
Capability sourceIterative editing and error correction
Vendor-statedBidirectional denoising is positioned for revision and correction workflows; the claimed quality remains vendor-evaluated.
Capability sourceResearch-only local generation
ChinaAPI inferenceThe official Transformers example is executable, but the lack of a vendor-confirmed production server keeps this below a normal commercial-serving recommendation.
Official launch / minimum
Evidence CTransformers path no hardware floor
The official Transformers generation path is executable, but the roughly 206GB BF16 repository and runtime overhead require operator planning; the vendor publishes no minimum GPU topology. Primary recipe
200-person team
Evidence EProduction server not confirmed
Do not size a shared service until a supported server path is validated against the actual diffusion steps, context and concurrency.
Commercial API
Evidence CNot ready for recommendation
The vendor card marks SGLang deployment support as coming soon. Wait for an official serving path and then validate redundancy and admission control. Primary recipe
Commercial-use check
Apache-2.0
MaaS / hosted service: No model-specific MaaS restriction identified in Apache-2.0.
Attribution: Provide the license and required notices, preserve attribution notices, and mark modified files; trademark rights are not granted.
Known limitations and open questions
- No ChinaAPI reproduction
- Vendor-confirmed SGLang deployment is still marked coming soon
- The active-parameter count and official minimum hardware topology are not published
- Diffusion decoding has different latency and batching behavior from autoregressive servers