Tavus Growth's 10 concurrent streams vs Sume's 4, 8 and 20 job slots

Tavus lists 10 concurrent streams on its Growth developer plan, 3 on Starter and 1 on Basic. Sume's generation concurrency is 1 on Free, 4 on Pro, 8 on Startup and 20 on Scale and Enterprise, and each plan also has a queue (Tavus pricing and Sume generation-admission docs, read 2026-10-09). The two numbers look alike but count different things: a Tavus stream is a live session, and a Sume slot is a render job in progress.

The limits side by side

Sume separates concurrency (jobs in processing) from queue capacity (accepted jobs waiting to start). The default queue is max(3, concurrency x 5). Accepted capacity is their sum.

What a limit means in practice

On a live product, the concurrent-stream limit is how many people can talk to an avatar at the same moment. If you plan an event where 40 people join together, a plan of 10 streams does not cover it as listed. The Tavus page does not say what happens to the 11th viewer, so ask before you rely on it.

On Sume, a full workspace does not reject the job. It accepts valid jobs as queued while queue capacity remains, and moves them to processing as slots free. A Pro workspace can hold 24 avatar jobs at once: 4 rendering and 20 waiting. Only the 25th returns 429 queue_full. The docs also say that concurrency comes from the plan and that prepaid top-ups do not raise it; the dashboard Concurrency tab is the source of truth.

Sizing a campaign

Suppose you need 60 personalized 20-second clips at plus ($4.90 each, $294.00 total). On Pro, submit them in waves of 24, wait for the queue to drain, then send the next wave. Three waves cover 60 clips (24 + 24 + 12). On Scale, 60 fits in one submission with room to spare, because accepted capacity is 120. In both cases the render happens in the background, and nobody has to be online while it runs.

If your need is simultaneous humans talking to an avatar, none of this helps, and the stream count is the number to check. If your need is a large batch of known messages, the accepted-job capacity is the number to check.

• Live sessions: ask Tavus what happens above the stream limit.
• Batches on Sume: submit up to accepted capacity, then wait for queued jobs to start.
• Treat 429 queue_full as a signal to wait, not to resubmit blindly.

Before you pick a plan

Read the effective limit from your own workspace, not from a table. Sume returns a generation_limits block on submit responses with the plan id, the concurrency limit and the remaining queue capacity, and the dashboard Concurrency tab is the source of truth. Check the Tavus plan page for the same on its side, and note that it does not describe queueing at all.

One more difference: Sume plan concurrency comes from the plan, so a larger balance does not raise it. If a campaign needs more slots for a week, the lever is the plan or an admin override, not a bigger top-up.