LLM Reference

The cheap leaderboard · for developers

Best for cheap

Decision area
Credible quality under $0.50 / 1M out.
Editor picks
4 editor picks
Eligible models
9 eligible models
See raw /best
EDITOR'S CHOICEResearched 8d ago

DeepSeek V4 Flash

DeepSeek · 1m context
Excellent

Embarrassingly good at near-zero cost — the new default for async work.

$0.12 / 1M out (live table $0.1173) with LiveCodeBench 91.6 and a 1M window, open source.

Open model
The numbers
$/1M out
$0.12
$0.06 input
Context
1m
max window
Pros
  • +$0.12 / 1M out
  • +LiveCodeBench 91.6
  • +1M context, open weights
Cons
  • −Lighter on long agentic chains

Also worth picking

The runners-up

ranked by editorial pick orderEditorial tiersExcellentStrongSolid
#2
Qwen3.5-Flash
Alibaba · 1m
$0.26 / 1M out
$0.26 out with a 1M window — superb for batch summarization and high-volume pipelines.
#3
Gemini 3.5 Flash-Lite
Google DeepMind · 1.05m
$2.50 / 1M out
Google's GA high-throughput pick: $0.30 / 1M input and $2.50 / 1M output, with a 1M context window, function calling, and structured outputs.
#4
GPT-5.4 Nano
OpenAI · 400k
$1.25 / 1M out
The only frontier-lab GPT that belongs on a cheap board — GPT-5.5 has no budget tier ($30 out); the nano is $1.25.

Eligibility

9 models are eligible for this board

Eligibility means tagged with useCases: [cheap]. Pins must come from this pool.
All picks