Best value LLM APIs
As of Oct 19, 2026, 5 of the 116 models with an LMArena score of 1350 or more are the cheapest way to reach their score. The top scorer, Gemini 4 Argon (1532), costs $2.00 / $10.00 per million input / output tokens; the cheapest of them, gpt-oss-120b (1366), costs $0.03 / $0.17.
A model is best value when no other model is both cheaper and higher scoring. Price is blended: five input tokens to one output token, at the model's cheapest seller. See also the cheapest models by quality tier.
Score against price
Blended price per million tokens (log scale) and LMArena score
Best value models (5)
Highest score first; each one down the list is cheaper and scores lower.
- 1
Google · top score
LMArena 1532
$2.00 / $10.00from Google Gemini
$3.33blended per 1M
- 2
Google · 2.7× cheaper, 33 points lower than the one above
LMArena 1499
$0.75 / $3.75from Google Vertex AI
$1.25blended per 1M
- 3
Xiaomi · 2.5× cheaper, 8 points lower than the one above
LMArena 1491
$0.43 / $0.87from DeepInfra, official $0.435 / $0.87
$0.5033blended per 1M
- 4
Z.ai · 4.8× cheaper, 20 points lower than the one above
LMArena 1471
$0.075 / $0.25from DeepInfra, official $0.15 / $0.50
$0.1042blended per 1M
- 5
OpenAI · 2.0× cheaper, 105 points lower than the one above
LMArena 1366
$0.03 / $0.17from W&B Inference
$0.0533blended per 1M
Scores are the LMArena text leaderboard's overall ratings (CC BY 4.0), which measure which answers people prefer; only the 116 models scoring 1350 or more with a fixed price are compared. Prices in USD per million tokens, from LiteLLM and models.dev (both MIT licence), checked daily. Routers such as OpenRouter are not counted as the cheapest seller: their price follows whichever host is cheapest that day. Confirm on the seller's pricing page before you rely on a number.