Model released in August now offered via API

"Motif 3," a language model from South Korean AI startup Motif Technologies, was listed on Nov. 2 (Nov. 3 Japan time) on U.S.-based OpenRouter, which lets users access multiple AI models through a single API. It is served by South Korean telecom giant KT, the only provider offering Motif 3 on OpenRouter. Whether Motif Technologies itself announced availability on OpenRouter had not been confirmed as of Nov. 11. The listing was reported on Nov. 11 by an X account that automatically detects and posts new models.

@gptbotio · November 11, 2026

Motif 3 is a model whose weights Motif Technologies released in August. It has 314 billion parameters in total and uses a "mixture of experts (MoE)" design that uses only 13.2 billion of them for each piece of text it processes. It mainly targets English and Korean, and its weights are published on Hugging Face under the MIT license.

Pricing at $0.30 input, $1.20 output

Pricing on OpenRouter is $0.30 for input and $1.20 for output per 1 million tokens. Reusing input that has already been read costs $0.06 per 1 million tokens. The context window is 262,144 tokens, and a single output can be up to 235,929 tokens.

On Oct. 30, KT officially launched "Token Factory," a platform that lets businesses use multiple AI models under a single contract, and outlined plans to link up with OpenRouter in November. KT serves seven models on OpenRouter, including Motif 3, KT's own "Mi:dm K 3.0 Pro" and NVIDIA's "Nemotron 3 Ultra," the same lineup as the seven models KT offers from its domestic facilities on Token Factory at a 30% discount. The discount is for businesses contracted with Token Factory and runs through Nov. 30.

No tool calling in the supported parameter list

In the OpenRouter listing, Motif 3's list of supported parameters does not include the ability to call external tools or programs (tools). It does support outputting the reasoning process and answering in a fixed format.

Motif Technologies positions Motif 3 as a model that is strong at tasks in which AI uses tools over long sequences of steps. In the company's own evaluation published on Hugging Face, it scored 76.2 on SWE-Bench Verified, which measures fixing software bugs, and 74.9 on Terminal-Bench 2.1, which measures terminal operation. In the same table, China's Alibaba "Qwen-3.7 max" scored 80.4 and 75.0, respectively.