ByteDance Opens Doubao Work to Third-Party Models: Kimi K3.1 Pro, GLM 5.3 and DeepSeek V4.1 Flash Join, With Per-Model Credit Multipliers and a Limited-Time 0.01x Rate for Doubao 2.1 Lite and DeepSeek V4.1 Flash

Background and Industry Pain Point

Chinese office agents are moving from in-house models only to one entry point with many models. Qwen Office, Yuanbao and others are adding outside models for a practical reason: different tasks suit different models, and no single in-house model balances quality, speed and cost everywhere. Doubao Work is ByteDance's AI agent product for office work and previously offered only its own Doubao 2.1 Lite and Doubao 2.1 Pro. On October 29 it opened third-party model configuration to team and enterprise subscriptions, billed against existing AI credits; on November 12 the capability was extended to a wider user base.

Architecture and Mechanism

All models are normalized into one credit system, and the model picker shows a credit multiplier for each: the lower the multiplier, the fewer credits the same task consumes. The first option is Auto (small tasks free), which picks a model by task size; for harder tasks users can switch manually to Doubao 2.1 Pro or a third-party model and adjust reasoning effort. Two limited-time offers ship alongside: credit discounts on participating models during designated weekday and holiday time slots, and reduced multipliers for Doubao 2.1 Lite and DeepSeek V4.1 Flash.

Benchmarks and Measured Results

No official success rates or benchmark comparisons for the models inside Doubao Work have been published. Multipliers from the UI screenshot and media reports: Doubao 2.1 Pro 0.65x, GLM 5.3 0.80x, Kimi K3.1 Pro 0.95x, Kimi K3.1 Pro high-speed 1.44x, Kimi K3 1.48x, GLM 5.3 Flash 0.08x, DeepSeek V4 Pro 0.46x, and Doubao 2.1 Lite and DeepSeek V4.1 Flash at a limited-time 0.01x. Running the same task on Kimi K3.1 Pro therefore costs about 1.5 times the credits of Doubao 2.1 Pro. Blogger @op7418 reported, as a personal test, editing an 8-minute video with DeepSeek V4.1 Flash and his own editing skill for credits worth about 0.12 yuan.

Developer Guide

Sign in to the desktop app at www.doubao.com/work and switch models from the Model menu above the input box. Use Auto or the two 0.01x models for everyday lookups and document cleanup, and move to Kimi K3.1 Pro or GLM 5.3 for long documents and harder reasoning while watching the multiplier gap. The 0.01x rate is a limited-time offer with no stated end date, so do not budget long-term costs on it. Teams already on Feishu and other ByteDance office tools will find adoption easiest.