hubai.uk Decision Room
Explore

Explore HubAI

NEWWhat AI should I use?Job + team size + goal → stack, risk and ROIBuild mine
HUBAI MODEL ROUTER LAB

Route the task.
Not the hype.

Tell HubAI what the model must do and what cannot be compromised. Get three candidates, a visible routing trace and the test required before deployment.

01 · ROUTING BRIEF

Define what should win.

Every input changes the shortlist. Nothing is hidden behind a chat response.

LIVE ROUTE

ChatGPT leads

ROUTE LEADER
TASKGeneral knowledge work→CONSTRAINTPublic or routine business work→ROUTEROUTE LEADER
01CG
Text AI

ChatGPT

100/100
Routing fit · v1.0

Broad multimodal work and a large product ecosystem

ACCESS / PRICE
Free; Plus and Team plans
WATCH
Capability and governance controls depend on the chosen plan
02GE
Text AI

Gemini

88/100
Routing fit · v1.0

Google Workspace, multimodal and search-adjacent workflows

ACCESS / PRICE
Free; Advanced plans
WATCH
Model names, access and controls can differ across Google products
PRIMARY TASK MATCH · DELIVERY MATCH · RISK ROUTE AVAILABLEOpen evidence
03CL
Text AI

Claude

84/100
Routing fit · v1.0

Long-document reasoning, synthesis and careful writing

ACCESS / PRICE
Free; Pro and Team plans
WATCH
Usage limits and tool availability vary by plan and region
SECONDARY TASK MATCH · PRIORITY MATCH · DELIVERY MATCHOpen evidence
ROUTING DECISIONChatGPT has the strongest rule-based fit. Verify it against Gemini before rollout.
02 · SAME-TASK PROOF

The router narrows the field. The pilot makes the decision.

Use one representative task, identical source material and one acceptance rule. Record full cost, latency, corrections and failures.

  1. 01Freeze the inputSame task, context and output format
  2. 02Measure the resultAccepted output, retries, time and cost
  3. 03Apply the risk gateData terms, access, oversight and failure route
  4. 04Record the winnerSave the evidence—not just the preference
Build the 14-day model pilot
FRONTIER WATCHLIST

New does not mean routed.

These launches remain outside the scored route until HubAI completes an independent same-task test.

A6
SCORE WITHHELD

GPT-6 Astra

Long, tool-heavy computer-use, coding, research and professional workflows

Why pausedHigh output-token priceReview launch evidence
F5
SCORE WITHHELD

Claude Fable 5.1

Long-running coding, research, knowledge work and context-heavy agents

Why pausedPremium $50/M output-token priceReview launch evidence
METHOD v1.0

Explainable by construction.

Task match contributes 50 points, priority 20, delivery 15, risk route 10 and budget access 5. The result is a context-fit route—not proof that one model is universally better.

Read HubAI methodology