LLM Reference

The summarization leaderboard · for knowledge workers

Best for summarization

4 editor picks · 5 eligible models · Faithful, structured, no hallucinated bullets.

See raw /best
EDITOR'S CHOICEResearched 5d ago

Gemini 3 Flash

Google DeepMind · 1M context
Excellent

Faithful, structured, cheap — and it won't invent bullets.

1M context plus MMLU-Pro 88.6 at $3 out — handles a 500-page transcript faithfully without breaking the budget.

Open model
The numbers
$/1M out
$3.00
$0.50 input
Context
1M
max window
Pros
  • +1M context for whole-doc summaries
  • +Faithful, well-structured
  • +$3 / 1M out
Cons
  • −Less nuanced than Opus on subtext

Also worth picking

The runners-up

ranked by editorial pick order
Editorial tiersExcellentStrongSolid
#ModelTier$/1M outEditor's note
#2
Claude Haiku 4.5
Anthropic · 200K
$4.00
Best faithfulness-per-dollar in the Claude line; doesn't hallucinate bullet points.
#3
DeepSeek V4 Flash
DeepSeek · 1M
$0.22
$0.22 out with a 1M window — the high-volume async summarizer.
#4
Claude Sonnet 4.6
Anthropic · 1M
$15.00
Highest-fidelity option when the summary is going in front of a customer.

Eligibility

5 models are eligible for this board

Eligibility means tagged with useCases: [summarization]. Pins must come from this pool.

All picks