LLMs under $20 per million tokens
Every model priced below $20 per 1M input tokens. Ranked by benchmark score within budget. The "I care about quality" tier.
Ranked by quality within budget
Input under $20/M, highest benchmark score first.
| # | Model | In $/1M | Out $/1M | Type |
|---|---|---|---|---|
| 1 | $5.00 | $30.00 | Closed | |
| 2 | $1.25 | $10.00 | Closed | |
| 3 | $0.39 | $2.34 | OSS | |
| 4 | $0.40 | $1.20 | OSS | |
| 5 | $1.25 | $10.00 | Closed | |
| 6 | $0.10 | $0.30 | OSS | |
| 7 | $0.09 | $0.29 | OSS | |
| 8 | $1.25 | $10.00 | Closed | |
| 9 | $1.10 | $4.40 | Closed | |
| 10 | $0.33 | $1.95 | OSS | |
| 11 | $1.75 | $14.00 | Closed | |
| 12 | $1.05 | $3.50 | OSS | |
| 13 | $0.60 | $6.00 | Closed | |
| 14 | $3.00 | $15.00 | Closed | |
| 15 | $0.26 | $1.00 | OSS | |
| 16 | $0.60 | $2.20 | OSS | |
| 17 | $0.00 | $0.00 | OSS | |
| 18 | $1.25 | $10.00 | Closed | |
| 19 | $0.03 | $0.14 | OSS | |
| 20 | $0.00 | $0.00 | OSS | |
| 21 | $0.08 | $0.40 | OSS | |
| 22 | $10.00 | $30.00 | Closed | |
| 23 | $0.30 | $0.50 | Closed | |
| 24 | $0.07 | $0.30 | Closed | |
| 25 | $0.30 | $1.20 | OSS | |
| 26 | $0.13 | $0.38 | OSS | |
| 27 | $0.10 | $0.78 | OSS | |
| 28 | $2.00 | $12.00 | Closed | |
| 29 | $0.25 | $2.00 | Closed | |
| 30 | $1.10 | $4.40 | Closed | |
| 31 | $0.20 | $0.80 | OSS | |
| 32 | $0.10 | $0.40 | Closed | |
| 33 | $0.32 | $0.89 | OSS | |
| 34 | $2.50 | $15.00 | Closed | |
| 35 | $0.78 | $3.90 | OSS | |
| 36 | $0.08 | $0.24 | OSS | |
| 37 | $1.00 | $3.00 | Closed | |
| 38 | $0.50 | $2.15 | OSS | |
| 39 | $0.54 | $0.54 | OSS | |
| 40 | $0.60 | $1.92 | OSS |
Top 3 best-quality sub-$20 models
GPT-5.5 sits in the under-$20 tier with a top-shelf quality score. Appropriate when reasoning or reliability matters and you do not need the absolute frontier.
GPT-5 Chat sits in the under-$20 tier with a top-shelf quality score. Appropriate when reasoning or reliability matters and you do not need the absolute frontier.
Qwen3.5 397B A17B sits in the under-$20 tier with a top-shelf quality score. Appropriate when reasoning or reliability matters and you do not need the absolute frontier.
The price gap · cheapest vs most expensive
At the top of this tier, pricing approaches frontier. The delta to true frontier (Opus, GPT-5) is 2x to 5x for roughly 5 to 15 percent quality gain on most tasks.