Open the model picker in Doubao Work and you will now find three outside models next to the in-house ones: Kimi K3.1 Pro, GLM 5.3 and DeepSeek V4.1 Flash. The November 12 update extends third-party models beyond team and enterprise plans to a wider set of users, and it ships with a limited-time promotion. The practical upshot: pick the wrong model and the same task can cost dozens of times more quota.
The multipliers: Flash and Lite at 0.01x for now
According to a breakdown on SMZDM citing the official announcement page, Doubao 2.1 Pro is listed at 0.65x, GLM 5.3 at 0.80x and Kimi K3.1 Pro at 0.95x. Doubao 2.1 Lite and DeepSeek V4.1 Flash both show 0.01x during the promotional period. These figures come secondhand, so trust what your own model picker shows.
Lower is cheaper.
Put plainly, running a job on DeepSeek V4.1 Flash during the promo deducts the least quota of anything on the list, while Kimi K3.1 Pro sits at the expensive end of the table, pricier even than Doubao's own 2.1 Pro.
If you would rather not choose, the official screenshot described by 17173 shows "Auto (small tasks free)" as the first item in the list, which matches a model to the size of the task. When a job gets harder, switch manually to Doubao 2.1 Pro or a third-party model and you pay that model's multiplier.
Light work on Flash, heavy work on 2.1 Pro first
Reformatting, summaries, bulk rewrites: send these to DeepSeek V4.1 Flash or Doubao 2.1 Lite. That is what they are built for.
Step up only when you need longer reasoning or harder writing. The gap between 0.65x and 0.95x is real, so try 2.1 Pro first and move to Kimi K3.1 Pro only if the output falls short.
There is a second promotion on paper too. Doubao Work says participating models will consume less quota during designated time slots on workdays and holidays. The cited official wording describes this as a planned promotion; the reports cited here give neither the hours nor the participating model list. Wait for the in-app notice before you schedule batch jobs around it.
The catch: no end date in the reports
The 0.01x rate is labeled limited-time, yet none of the reports cited here give an end date. Don't build a whole workflow on it, because when the multiplier climbs back, your quota will go faster.
Company accounts actually got this first. IT Home reported on October 29 that third-party models opened to team and enterprise subscribers, with usage deducted from the organization's existing AI points and admins able to enable models in bulk by vendor. If your work account doesn't show them, ask your admin.
On the personal side, the personal plan dropped its 5-hour short-cycle quota limit on November 7, so usage now only counts against a weekly allowance. Keep two costs apart, though: the subscription fee stays the same, and a multiplier only sets how fast quota drains. 0.01x is not a price in yuan. The available numbers let you compare models against each other, not estimate what a month of heavy use will cost.
Next step: confirm the third-party models and their multipliers appear in your account, then run one identical task on Flash and on 2.1 Pro. Compare the output and the quota deducted before you settle on a default.