GitHub Copilot Model Multipliers & Best Usage Strategy (Pro+)

Cheap models (0.33×):
Haiku 4.5, Gemini 3 Flash, GPT‑5 mini, Raptor mini, MAI‑Code‑1‑Flash
→ Use these as your main workhorses.

Moderate (1×–3×):
Gemini 2.5 Pro (1×), GPT‑5.1‑Codex (3×)
→ Use only when you need deeper reasoning.

Expensive (6×–14×):
Sonnet 4.5/4.6, Gemini 3 Pro/3.1 Pro, Gemini 3.5 Flash, GPT‑5.3/5.4
→ Avoid unless absolutely necessary.

Extremely expensive (27×–57×):
Claude Opus (all versions), GPT‑5.5
→ Avoid completely.


My cost‑efficient model workflow

  • Gemini 3 Flash (0.33×) → everyday dev work
  • Claude Haiku 4.5 (0.33×) → documentation & structured text
  • GPT‑5 mini / Raptor mini (0.33×) → repetitive coding
  • GPT‑5.1‑Codex (3×) → deep reasoning & complex debugging
3 Likes

Very useful for Copilot Users Ben! That new regime of theirs is daylight robbery. Absolutely no point in having a Copilot subscription now as essentially zero benefit cost-wise (better off with Open Router). Also Copilot still does not allow using Kimi, DeepSeek, MiniMax, or others to aid in decreasing those costs. Pretty silly of them to hook you in to a fixed selection. Besides the performance of Claude has seriously decreased over recent weeks (personal opinion of course), seems like it loves to digest tokens like they are going out of fashion too, even on lower reasoning levels increasing consumption even faster, combined with the new multipliers it is a loss/loss situation (outside of your above suggestion to compensate for it by avoiding Claude if the User do without it)... Bar stewards got you by the short and curlies.