Your Copilot Chat Can Run Multiple Models Now

GitHub's multi-model research preview is now in the Copilot Chat model picker. It is not Auto with a new coat of paint. Single, Cascade, and Critique pick how many models run a turn.

News · No. 32 · Wed Oct 7

HydraFusion Moved Into VS Code

GitHub's multi-model research preview is now in the Copilot Chat model picker. It is not Auto with a new coat of paint. Single, Cascade, and Critique pick how many models run a turn.

Multi-model routing landed in the editor you already pay for. Devin got cheaper the same week. Google's next frontier model is still cyber-gated. Here is the part that changes Monday.

In This Email

» HydraFusion in VS Code and the Copilot app: workflows, not Auto 2.0

» Devin 30-40% cheaper, Gemini 4 Argon cyber-gated, Microsoft streaming STT #1

» Skip calling HydraFusion “Auto 2.0.” Auto picks a model. HydraFusion picks a workflow.

The Big Thing: Multi-Model Inside the Editor You Already Use

HydraFusion Is in VS Code and the Copilot App

On September 30 GitHub put HydraFusion in VS Code 1.140+ and the Copilot app. It sits in the Copilot Chat model picker as a research preview. It is not one model. HydraFusion selects Single, Cascade, or Critique, and that choice decides how many models run on a turn.

Why you care: you can test multi-model orchestration inside the editor you already pay for. No new agent seat. No context switch to a separate CLI for a first look. Copilot Pro, Pro+, Business, and Enterprise all see the same picker row.

What Single, Cascade, and Critique actually do:

Single sends the turn to one selected model and stops. Fastest path. Narrow asks: rename a helper, explain a stack trace, draft a short commit message.

Cascade starts with an efficient model, then a quality gate accepts that draft or escalates to a stronger model. Cheap-first with a bailout when most asks are routine but a few need the expensive pass.

Critique has one model draft, an independent read-only critic from a different model family reviews it (same pattern as Rubber Duck), then the drafting model revises once. Use it when correctness beats speed - a risky refactor, a migration plan, anything you would ask a teammate to second-pair.

Monday use I would run: open Copilot Chat, select HydraFusion, paste the PR description for a change you plan to ship this week, and ask it to flag merge risks and missing tests. Watch which workflow it picks. Run the same ask under Auto. Write the difference in one sentence for your team channel.

If it does not appear in VS Code, enable chat.copilot.hydraFusion.enabled. On Business or Enterprise, an admin may need to allow preview features. In the Copilot app: update, search Settings for HydraFusion, turn it on, pick it from the model list.

My take: treat Single / Cascade / Critique as three workflows, not three model names. Auto still picks a model per request. HydraFusion picks a workflow and can coordinate multiple models in one turn. Write that distinction on a sticky note before you ship a prompt that assumes they are the same. This release's progress updates make that easier to see mid-turn.

Worth Your Time

Devin Got 30-40% Cheaper

Cognition cut Devin usage cost: 30-40% in Fusion/Normal, 15-20% in Ultra, up to 70% in Devin Review. Fusion still leads FrontierCode 1.1 Extended at 68.8 (~$0.60/task). If you budget agent hours weekly, the same seat now buys more merged work before you rethink the tool.

What the same seat now buys: more Fusion/Normal tickets per dollar, cheaper Review passes you can leave on by default, and less of a budget cliff when Ultra is the right call. Re-run last month's agent spend against the new rates before the next seat renewal. Devin's write-up.

Gemini 4 Argon Is Real - and Still Cyber-Gated

Google announced Gemini 4 Argon on Sep 30 with 1M output tokens and cyber-defense training. Rollout is Fairwind Program plus voluntary US pre-release access, not a public API yet. Intro API pricing named: $2/$10 per 1M in/out (then $4/$20).

What you can do now: read the price card, note the cyber-first gate, and decide whether to apply for Fairwind or wait for public API. Bookmark the post so finance is not surprised when rates go live.

What you cannot do until public API: swap Argon into Monday production prompts, wire it as your app's default model, or treat intro rates as a guarantee for every workload. Pre-release is not a drop-in. You cannot switch Monday - but the price card and cyber-first gate still show how Google will ship the next frontier wave. Google's post.

Microsoft's Streaming STT Hit #1

MAI-Transcribe-2-Streaming ranks #1 on Artificial Analysis for final and partial transcripts. First partials in just over 100ms. Intro $0.54/hour through year-end. Ships with MAI-Voice-2.1 (23 languages) and Flash. If you are wiring voice agents, the transcription half of the loop just got a public price and latency target you can design against. Microsoft AI's write-up.

Skip It

Calling HydraFusion “Auto 2.0”

GitHub says Auto picks a model per request. HydraFusion picks a workflow and can coordinate multiple models in one turn. Same picker row. Different job. Do not swap the labels in your runbook.

Every check here gets run in the free AI Academy.

Join the free AI Academy →

Free, no card. The tools and people behind this issue live here.

That's the Debrief.

See you Friday for the tools.

From Drew and The AI Debrief team

P.S. Did you try Single, Cascade, or Critique first in HydraFusion? Hit reply. We read every one.

P.P.S. New here, or skipped one? Every past edition is in the archive.