LLM Reference

The docs q&a leaderboard · for knowledge workers

Best for docs q&a

4 editor picks · 7 eligible models · Grounded answers over your own knowledge base.

Editorial pick plus benchmark and API pricing context.

See raw /best
EDITOR'S CHOICEResearched 24d ago

Claude Sonnet 4.6

Anthropic · 1m context
Excellent

The safest grounded-QA pick in production.

Highest groundedness and graceful refusal — says "not in the docs" instead of guessing.

Open model
The numbers
$/1M out
$15.00
$3.00 input
Context
1m
max window
Pros
  • +Strong groundedness
  • +Refuses cleanly off-context
  • +1M context
Cons
  • −$15 / 1M out

Also worth picking

The runners-up

ranked by editorial pick orderEditorial tiersExcellentStrongSolid
#2
Gemini 3 Pro
Google DeepMind · 1m
$5.00 / 1M out
A 1M window lets you skip retrieval entirely on medium corpora.
#3
Claude Haiku 4.5
Anthropic · 200k
$4.00 / 1M out
Almost as grounded as Sonnet at a fraction of the price — the default for chat-on-docs.
#4
DeepSeek V4 Pro
DeepSeek · 1m
$0.87 / 1M out
Open weights, 1M context, cheap — strong self-hosted RAG backend.

Eligibility

7 models are eligible for this board

Eligibility means tagged with useCases: [docs-qa]. Pins must come from this pool.

All picks