DeepSeek folds three modes into V4.1 Flash as V4 Pro bows out

DeepSeek has launched V4.1 Flash as a unified model that automatically chooses between fast, expert and image-recognition capabilities. The change matters because users no longer need to select modes manually, while V4 Pro is being retired and replaced by the newer model’s pricing.

DeepSeek Watch analysis

What happened

AIBase reports that DeepSeek’s new model can detect query complexity, activate visual capabilities when an image is supplied and adjust its processing effort accordingly. The report says V4 Pro was scheduled to be retired at 12:00 Beijing time on 14 September 2026, with requests routed to V4.1 Flash and charged at the newer rates.

The reported Flash-series pricing from 10 September was 0.02 yuan per unit for cached input during off-peak hours, 1 yuan for cache misses and 4 yuan for output. Peak-hour prices were reported as double those rates.

Key findings

• One unified mode
  DeepSeek says users no longer need to switch manually between fast, expert and image-recognition modes.
• Automatic visual handling
  The model can reportedly activate visual capabilities when it detects image input.
• V4 Pro retirement
  AIBase reports that V4 Pro was routed to V4.1 Flash from 14 September.
• New pricing
  Reported off-peak prices are 0.02 yuan for cache hits, 1 yuan for cache misses and 4 yuan for output.

Why it matters

For developers, the practical promise is simpler model selection: one endpoint can decide how much computation and which capabilities a request needs. That could make mixed text-and-image workloads easier to operate, although the supplied evidence does not independently verify DeepSeek’s claims about performance, cost or speed.

The pricing change also makes the migration from V4 Pro a procurement decision, not merely a model-name shuffle. Teams using cached prompts or long-running workloads will need to check which rate applies at the time of use.

Our read

This is a meaningful product consolidation, with a bill attached. DeepSeek users should review V4 Pro integrations, confirm the replacement model’s pricing and test image-heavy workloads before assuming the new automatic mode is a free upgrade.

What to watch

• Whether independent tests confirm the claimed performance and cost advantages over V4 Pro.
• How reliably V4.1 Flash chooses processing depth across mixed workloads.
• Whether DeepSeek publishes clearer migration guidance and current regional pricing.
• The effect of peak-hour pricing on high-volume API users.

Discussion spark: Would automatic mode selection make you more likely to use one general model, or do you still want explicit control over reasoning and vision?

not affiliated with or endorsed by DeepSeek