Its weights are available under Apache 2.0.
The company says Flint supports up to 512k tokens of context, while recommending use within 256k. It lists compatibility with llama.cpp, Ollama, Unsloth, LM Studio, vLLM, and SGLang, and has also released rec, an inference engine tuned for Flint on M4 Pro, M5 Pro, and M5 Max Macs.
Recursive reports more than 125 tokens per second on an M4 Pro MacBook Pro and 150 tokens per second on a 64GB NVIDIA RTX Spark at short contexts. It says Flint leads sub-50B MoE models on SWE-Bench Pro v2, Terminal-Bench 2.1, and SWE-Bench Verified. The source does not establish independent benchmark replication, comparative performance under matching settings, or performance on longer contexts.