by InclusionAI
Ring-flash-2.0 is a high-performance thinking model deeply optimized based on the Ling-flash-2.0-base. It uses a mixture-of-experts (MoE) architecture with a total of 100 billion parameters, but only activates 6.1 billion parameters per inference. The model employs the original Icepop algorithm to solve the instability issues of large MoE models during reinforcement learning (RL) training, enabling its complex reasoning capabilities to continuously improve over long training cycles. Ring-flash-2.0 has achieved significant breakthroughs on multiple high-difficulty benchmarks, including mathematics competitions, code generation, and logical reasoning. Its performance not only surpasses top dense models under 40 billion parameters but also rivals larger open-source MoE models and closed-source high-performance thinking models. Although the model focuses on complex reasoning, it also performs exceptionally well on creative writing tasks. Furthermore, thanks to its efficient architecture, Ring-flash-2.0 delivers high performance with low-latency inference, significantly reducing deployment costs in high-concurrency scenarios.
Ring-flash-2.0 is a high-performance thinking model deeply optimized based on the Ling-flash-2.0-base. It uses a mixture-of-experts (MoE) architecture with a total of 100 billion parameters, but only activates 6.1 billion parameters per inference. The model employs the original Icepop algorithm to solve the instability issues of large MoE models during reinforcement learning (RL) training, enabling its complex reasoning capabilities to continuously improve over long training cycles. Ring-flash-2.0 has achieved significant breakthroughs on multiple high-difficulty benchmarks, including mathematics competitions, code generation, and logical reasoning. Its performance not only surpasses top dense models under 40 billion parameters but also rivals larger open-source MoE models and closed-source high-performance thinking models. Although the model focuses on complex reasoning, it also performs exceptionally well on creative writing tasks. Furthermore, thanks to its efficient architecture, Ring-flash-2.0 delivers high performance with low-latency inference, significantly reducing deployment costs in high-concurrency scenarios.
On AIHubMix, inclusionAI/Ring-flash-2.0 costs $0.14 per million input tokens and $0.54 per million output tokens.
inclusionAI/Ring-flash-2.0 accepts text input.
inclusionAI/Ring-flash-2.0 supports thinking, tool calling, function calling and structured outputs. Per-protocol parameter support is listed in the capability table on this page.
inclusionAI/Ring-flash-2.0 is available through the AIHubMix unified API. The API is OpenAI-compatible: point your OpenAI SDK at https://aihubmix.com/v1, use your AIHubMix API key, and set the model name to inclusionAI/Ring-flash-2.0 — no other code changes needed.
inclusionAI/Ring-flash-2.0 is developed by InclusionAI. AIHubMix aggregates it alongside models from other providers behind one API and one bill.
Ling-1T is the first flagship non-thinking model in the “Ling 2.0” series, featuring 1…
Ring-1T is an open-source idea model with a trillion parameters released by the Bailing…
Ling-flash-2.0 is a language model from inclusionAI with a total of 100 billion…
Ling-mini-2.0 is a small-sized, high-performance large language model based on the MoE…
Use inclusionAI/Ring-flash-2.0 via the AIHubMix unified API — one interface for every major LLM.