DeepSeek V4.1 Flash — On Recommended, Budget 1×

September 15, 2026

DeepSeek V4.1 Flash is in the picker: a 1M-token context, tools, and zero-retention hosting, at Budget 1×. Auto can route to it, and you can pin it from Recommended when you want DeepSeek on every turn.

What you can do

• Pick DeepSeek V4.1 Flash from Recommended, or from Fast. It is DeepSeek's first Causal Encoder–Decoder Flash: 552B total, 8B active on input / 16B on output, reasoning and tools, with 1M context and 65K max output.
• Let Auto pick it. V4.1 Flash is in the workhorse Auto pool at quality 60, so Auto can route to it on Community and paid plans. It is also Swarm-eligible.
• Stay on zero-retention hosts. Every request is pinned to OpenRouter endpoints that keep nothing and can still call tools.
• Still pick Gemini 3.8 Flash from Fast. It left Recommended only; Auto, Swarm, and pinned Gemini sessions are unchanged.
• See Claude Opus 5 on Recommended. Anthropic's flagship now sits on that strip with GPT-6 Astra and GLM 5.3, instead of being ranked out of the shortlist.

Where this shows up

You open a new chat and the landing card says Try out the latest DeepSeek V4.1 Flash — Try it today. Click it and the picker opens with that row highlighted.

You leave the model on Auto. On a workhorse turn, Alfrada OS can now pick DeepSeek V4.1 Flash — quality 60, Budget 1×, ZDR-pinned.

You want Flash-class coding or computer-use at Budget 1×, without leaving zero-retention hosting. Pin DeepSeek V4.1 Flash instead of hunting under Fast.

You reach for Anthropic's flagship from Recommended and Claude Opus 5 is actually there.

Try it

• From the empty-chat card, or pick DeepSeek V4.1 Flash in Recommended, then: "Refactor this module across files, keeping tool use tight and the reasoning visible."
• "Inspect this repository, plan the refactor, implement it, and verify the risky paths."
• Pick Claude Opus 5 from Recommended, then: "Challenge this market-entry plan, research the missing evidence, and produce a board-ready recommendation."

Heads up

• Budget 1× — that is the token band. Community accounts that are capped at 1× can pick it, and Auto can route to it.
• Auto-eligible. Quality 60 sits on the paid Auto floor; the workhorse tier means paid Auto will consider it, not skip it as a budget-tier model.
• Recommended is seven models: GPT-6 Astra, Claude Opus 5, GLM 5.3, GPT-5.6 Sol, Kimi K3, Grok 4.6, DeepSeek V4.1 Flash. Gemini 3.8 Flash stays under Fast.
• Older DeepSeek V4 pins that already rolled to Gemini 3.8 Flash stay there until you switch.