RealRun / Guides

Best LLM for email triage (2026)

Short answer: On 30 fresh, held-back emails, GPT-6.1 Sol and the low-cost GPT-5.4 mini both scored 95%. Claude Opus was close at 94.4%. We don't recommend Claude Haiku or Grok 4.20 for inbox triage: both filed scam emails as "Act" in some passes.

Scores

Each email sorted into Act, Read later or Archive, with scams caught. Accuracy % on 30 held-back emails and on 50 main emails, average of passes.

Model	Held-back 30	Main 50	Scam miss?	Cost
GPT-6.1 Sol	95.0%	95.0%	No	medium
GPT-5.4 mini	95.0%	93.0%	No	low
Claude Opus	94.4%	96.0%	No	subscription
Claude Sonnet	91.7%	94.0%	No	subscription
Gemini 3.8 Flash	91.1%	94.0%	No	low
Grok 4.7	90.0%	92.0%	No	high
Grok 4.20	86.7%	92.0%	Yes ⚠	low
Claude Haiku	85.6%	86.0%	Yes ⚠	subscription

What we found

Cheap can be good here: GPT-5.4 mini matched the top score at a low cost.
Scams are the real risk. Claude Haiku filed gift-card and payment-request scams as Act in some passes; Grok 4.20 did it on the fresh set in both passes.
Scores moved between passes, so we ran each set at least twice.

Full inbox triage results (80 emails, 1,420 graded answers).

Want the exact prompt and setup? The RealRun Inbox Triage Playbook (v2) has the word-for-word tested prompt, model picks by budget, and every miss explained.

Get the playbook, $29 or get the free weekly AI test report

More from RealRun

Scores reflect our tests on the dates shown, not a guarantee. Claude models ran on a Claude subscription, so they have no metered API cost. cost_tier is relative within a task (low/medium/high) from our run spend; 'subscription' = not metered. Tested October 2026 by RealRun (Base Ledger IP LLC).