LLM Reference

LLaDA2.2 Models by InclusionAI

InclusionAIApache 2.0Open sourceOpen Source
2 models2026Up to 131k ctx

Last refreshed 2026-09-07. Next refresh: weekly.

Details

ResearcherInclusionAI
LicenseApache 2.0OSI-approved
Commercial useCommercial use: permitted
Models2
Released2026
Max context131k

Capabilities

JSON / Tool useAll models

Links

WebsiteHuggingFace

About

LLaDA2.2 is InclusionAI's agentic MoE diffusion language-model series with Levenshtein Editing (DELETE/INSERT control tokens), Block Routing for long-context MoE diffusion, and L-EBPO agentic RL. First-party Hugging Face ships LLaDA2.2-flash (100B non-embedding) and LLaDA2.2-mini (16B / ~1.4B active), both with 128K context under Apache 2.0. Tech report and LICENSE live in github.com/inclusionAI/LLaDA2.X. Compare it for Agents, Coding, and Long context.

Decision facts

Best fit
agenticagentcoding
Capability starting point
LLaDA2.2 Mini with 131k context and JSON / Tool use
Lowest tracked input
Not tracked
Closest related family
K2 Horizon

Current Variants

Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.

2 in view
LLaDA2.2 MiniCurrent

Use when the workload needs agentic, 131k context, and JSON / Tool use.

2026-09agentic131k contextJSON / Tool use
LLaDA2.2 FlashCurrent

Use when the workload needs agentic, 131k context, and 100B parameters.

2026-07agentic131k context100B parameters

Release Timeline

2 release groups
2026-09
1 current
LLaDA2.2 Mini
agentic131k contextJSON / Tool use
Current
2026-07
1 current
LLaDA2.2 Flash
agentic131k context100B parameters
Current

Specifications(2 models)

LLaDA2.2 model specifications comparison
ModelReleasedContextParametersJSON / Tool use
LLaDA2.2 Mini2026-09131k16B (1.4B active)Yes
LLaDA2.2 Flash2026-07131k100BYes