ByteDance now prices rival models inside its office agent through one credit system, giving the makers of Kimi, GLM and DeepSeek a channel while ByteDance sets the meter.
ByteDance has added Kimi K3.1 Pro, GLM 5.3 and DeepSeek V4.1 Flash to Doubao Work, its AI agent for office tasks, widening access after a first release to business subscribers on October 29. IT Home reported the update on November 12 from an official Doubao announcement, which also brought in a metering system: the model picker now shows how fast each model draws down a user’s quota, and two limited-time discounts lower that rate.
Doubao Work offered only ByteDance’s own Doubao 2.1 Lite and Doubao 2.1 Pro before the update, according to IT Home. Users can now switch between in-house and outside models task by task. The first item on the list is "Auto (small tasks free)", which matches a model to the size of the job; heavier work can be moved by hand to Doubao 2.1 Pro or a third-party model and billed at that model’s multiplier.
None of the reports give the multiplier values, the discount rates, the hours the discounts apply or when the offers end. A lower multiplier means the same task consumes less quota, The Paper explained, so the figures that matter for cost comparisons are still visible only inside the app.
ByteDance tested the models with paying teams first
The November 12 update widens a feature that started with business customers. On October 29 Doubao Work switched on the three outside models by default for its team and enterprise subscription plans, Mydrivers reported in a piece carried by Sina Tech. Companies did not need to buy access from each model maker; usage came out of the AI points already in the company account. Administrators can grant models to individuals, departments, groups or all staff, and limit them to Doubao’s desktop, mobile and web apps or to Doubao Work inside Feishu. Mydrivers said then that the personal edition did not have the models and that more third-party options were planned.
The two discounts now in place are a cut in quota use for participating models during set periods on workdays and holidays, and a temporary reduction in the multipliers for Doubao 2.1 Lite and DeepSeek V4.1 Flash. National Business Daily added that new users who sign in to the desktop version receive 30 days of subscription benefits free. Geekpark said all models run on one unified credit system, which lets users compare costs directly. It described Kimi as built for very long documents of about one million characters, GLM 5.3 for Chinese writing and comprehension, and DeepSeek V4.1 Flash for quick, low-cost daily tasks.
Rival office agents already let users pick outside models
ByteDance arrived late to this. The Economic Observer, in a report republished by Phoenix on November 5, noted that Doubao Work launched on August 25 by absorbing TRAE Work, Coze and Feishu Aily, products that had let users choose third-party models, yet it shipped with Doubao models only. Tencent’s WorkBuddy has supported DeepSeek, GLM, Kimi, MiniMax and Tencent’s own Hunyuan since launch. Alibaba’s Qwen Office added GLM 5.3 and DeepSeek V4 Pro in mid-August, then replaced model names with tier labels. On October 29, Xiaomi’s MiMo desktop app added seven third-party models for members.
The move comes as Doubao Work trails its main rival in usage. In September, WorkBuddy had 32.353 million monthly active users, ahead of Baidu’s Dazi and Alibaba’s Qwen Office, with Doubao Work outside the top three, according to QuestMobile figures EastFrontier reported when WorkBuddy led China’s AI office market. The Economic Observer quoted a developer who said offering business customers more model choice was close to inevitable, because models differ in strengths and tiering them by job can help companies control token costs.
Quota multipliers will show how ByteDance prices rivals
The three models carry very different list prices outside Doubao Work. The Economic Observer gave output prices per million tokens of 68 yuan for Kimi K3.1 Pro and 28 yuan for GLM 5.3, and 4 yuan off-peak or 8 yuan at peak for DeepSeek V4.1 Flash. Doubao Work declined to tell the paper on November 4 why it made the change or whether the personal edition would get third-party models.
The multipliers ByteDance sets against those price gaps, and whether they hold once the limited-time cuts end, will show whether Doubao Work treats outside models as a convenience or as a way to steer users back to Doubao 2.1. A further batch of third-party models, which Mydrivers said was planned, would show how open ByteDance intends the product to be.