Mergestorm’s review fleet moves to DeepSeek V4.1 Flash

DeepSeek shipped V4.1 Flash on September 10, 2026. The official model id is deepseek-flash. Mergestorm’s review fleet now uses that generation for the DeepSeek work: Vortex on every push, Tempest when you pick the cheap deep-review slot, Work chat, and the short classify / house-rules calls.

The product name in Agents is DeepSeek V4.1 Flash. A Tempest run on that slot costs 4 credits. Vortex is still 1 credit. We did not turn the whole fleet into a large-model pass, and we did not change Cyclone’s patch model.

What moved

Three things had to change on our side. The API endpoint did not.

Accounting. Observability bills every DeepSeek id at the V4.1 Flash list. Off-peak uncached input is $0.15 / 1M tokens, cache hits $0.003, output $0.60. Peak is 2× on weekdays 01:00–04:00 and 06:00–10:00 UTC. Weekends are off-peak.

Tempest. The DeepSeek row in the orchestrator picker is V4.1 Flash at 4 credits. It used to be V4 Pro at 6. If a saved setting still says deepseek-v4-pro, we still accept it and charge 4. DeepSeek retires that id on September 14, 2026, 04:00 UTC and then routes it to Flash anyway. We are not listing Pro and Flash side by side. After the 14th they would be the same model at two prices.

Classify. Comment classification and house-rules stay on Flash, with thinking off. Those are 64-token one-shots. Tempest exploration and Vortex tool loops still think.

Vortex Core and the specialist lanes were already on the Flash alias. This launch is the rest of the DeepSeek surface catching up, plus the credit and rate-card truth.

What did not change

Vortex still reviews every push. Vortex V2 is still the frequent, convergent loop. Cyclone still patches clear findings; the patch model remains Luna. Tempest is still the on-demand deep pass, and the default orchestrator is still Opus.

We are not publishing a Mergestorm quality bake-off for V4.1. DeepSeek’s own launch post is the source for how Flash compares to Pro. The August Vortex study of Haiku 4.5 vs the previous Flash generation is a different model and a different question. Do not read it as a V4.1 result.

Credits

A credit is the same unit it was yesterday. The DeepSeek Tempest price is the part that moved.

Free is still 30 credits a month, so a free account can run several Flash Tempest reviews in a month, or one Astra review. Paid piles did not change with this cutover. How those piles were set is in our review credits post. Live numbers are on Billing.

Astra is the other recent Tempest addition. If the uncertainty lives between files and you want the more expensive orchestrator, pick Astra. Flash is the DeepSeek slot for the same Tempest loop when you want the cheap, fast deep pass.

How to run Flash

Open Agents, select DeepSeek V4.1 Flash, and run Tempest on the pull request. The GitHub comment names the orchestrator that ran.

Use Flash when the PR needs a wider read than Vortex and you want it at 4 credits. Use Astra, Sol, or Fable when the failure mode is expensive enough to pay for a frontier orchestrator. Leave Vortex on for every push either way.

Questions

Do I have to change a setting? No, if you already used the DeepSeek Tempest slot. The picker label is now DeepSeek V4.1 Flash. A leftover Pro selection still runs and still costs 4 credits.

Is Vortex more expensive? No. Vortex stays 1 credit. The 4-credit price is Tempest DeepSeek only.

Did Cyclone change? Classification did. The patch model did not. Luna still writes the patch.

Where is the official model card? DeepSeek’s V4.1 Flash announcement and API pricing.

Pick Flash. Keep the fleet moving.