InclusionAI / inclusionai/ling-3.0-tiny

Ling 3.0 Tiny - access through LLMTR

Ling 3.0 Tiny is InclusionAI's small Mixture-of-Experts model: only 1.3 billion of its 7.9 billion parameters run per token. It has two modes. In the default Thinking mode it reasons step by step before answering and returns that chain separately in `reasoning_content`. In Instant mode it answers directly; switch to it with the `:fast` suffix or by sending `"reasoning": false`. It supports native tool/function calling, reads from a prompt cache on repeated prefixes, and offers a 256K context window. A single response can be at most 32,768 tokens. It accepts text only — no image or audio input, and no schema-enforced JSON output. In Thinking mode the reasoning tokens count as output and are returned inside `completion_tokens`; the model can think at length even on short questions, so keep `max_tokens` generous or switch to Instant mode when you need speed.