Quality per dollar.
Cheapest capable AI models.
Composite of MMLU and HumanEval divided by per-million input-token API price. Frontier models cost a lot; this list surfaces the cheapest options that still hold up on the basics.
Benchmarks used:
MMLU · 50%
HUMANEVAL · 50%
$ weighted
Score divided by the model's per-million input-token API price. Open-weight models without published API pricing are excluded.
| # | Model | Score | From |
|---|---|---|---|
| 1 |
84.6
$0.15/M in
|
OpenAI | |
| 2 |
85.6
$0.27/M in
|
DeepSeek | |
| 3 |
90.0
$0.55/M in
|
DeepSeek | |
| 4 |
80.5
$1.0/M in
|
Anthropic | |
| 5 |
86.3
$1.25/M in
|
Google DeepMind | |
| 6 |
81.9
$1.25/M in
|
Google DeepMind | |
| 7 |
88.0
$2.0/M in
|
Mistral AI | |
| 8 |
92.1
$2.5/M in
|
OpenAI | |
| 9 |
89.5
$2.5/M in
|
OpenAI | |
| 10 |
90.4
$3.0/M in
|
Anthropic | |
| 11 |
88.8
$3.0/M in
|
Anthropic | |
| 12 |
88.4
$3.0/M in
|
xAI | |
| 13 |
72.7
$2.5/M in
|
Cohere | |
| 14 |
86.4
$10.0/M in
|
OpenAI | |
| 15 |
90.0
$15.0/M in
|
Anthropic |
Showing top 15 models with published data on at least one of the benchmarks above. Scores are weighted averages on a 0–100 scale.