DeepSeek Flash: Thinking and Effort Levels Come to DeepSeek

DeepSeek has moved on from the model names we launched with. Since the V4 release, the current model is addressed as DeepSeek Flash, currently serving V4.1-Flash, and the old “deepseek-chat” and “deepseek-reasoner” names are no longer part of DeepSeek’s model list. AI Data Stream now follows suit, and brings the effort control that Claude and GPT users already have to DeepSeek.

What Changed for You

Nothing you need to do. DeepSeek Flash is already the DeepSeek model in the picker.

• Chat and Reasoner are retired. DeepSeek dropped both names with V4 and has been serving them from the same Flash model since. Anyone still on either entry is migrated to DeepSeek Flash automatically.
• Thinking is on. DeepSeek Flash reasons before it answers, which is what the old Reasoner entry did. In our own comparison on an analytics question, the thinking answers computed the click-through-rate drop and split branded from non-branded queries, details the non-thinking answer skipped.
• You control how much. The chat now has an effort picker for DeepSeek with three levels: Low, High and Max. Team owners set the default under Settings > AI Providers, and anyone can change it per message in chat, exactly as with Claude and GPT.

Your API key, settings and conversations all remain the same.

Pricing

As always, AI Data Stream is bring-your-own-key: DeepSeek bills you directly at its published rates, which depend on the time of day. Off-peak rates are half the peak rates.

• Off-peak: $0.15 per million input tokens (cache miss), $0.003 for cache hits, $0.60 per million output tokens.
• Peak: $0.30 per million input tokens (cache miss), $0.006 for cache hits, $1.20 per million output tokens.
• DeepSeek defines peak hours as 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday through Friday, excluding Chinese public holidays. All other hours, weekends and those holidays are off-peak.

Thinking tokens count as output. In our comparison, a High-effort answer used roughly six times the output tokens of the same answer without thinking, so pick Low for quick questions and save Max for the analysis that matters.

What DeepSeek Flash Offers

According to DeepSeek’s documentation, DeepSeek Flash has a 1M-token context window, up to 384K output tokens, and supports tool calls, JSON output and vision. Thinking is the default mode. Every AI Data Stream analysis is a chain of tool calls against your connected data, and the model handled our multi-step tool tests, including a streamed two-tool exchange at Low effort.

Switching Models

The picker now offers a single DeepSeek model, so there is nothing to choose there. To set your team’s default effort:

1. Go to Settings > AI Providers
2. Click your DeepSeek provider
3. Pick a Reasoning Effort
4. Save

Or simply pick the effort from the picker next to the model when starting a conversation.

Questions?

If you have any questions about the transition, contact us or check our AI providers documentation.