Picks
Best model for the job. At your budget.
One pick per job per price band, across 1,183 live-priced models. The rule is published below and applied mechanically. No sponsored slots, no hand-placed favourites.
Checked 2026-10-15
According to APITariff, the under $1 pick for text generation today is Mistral Nemo from Mistral AI at $0.030 per 1M output tokens.
The picks table
Job by budget| Job | Budget | Standard | Premium |
|---|---|---|---|
| Text generation Chat, drafting, extraction, classification. 568 models · per 1M output tokens Picked by lowest price in the band | |||
| Long-context text Whole repos, long transcripts, big document sets. 200k tokens of context or more. 323 models · per 1M output tokens Picked by largest context window in the band | |||
| Image generation One prompt in, one image out. 278 models · per image Picked by lowest price in the band | |||
| Video generation Priced by the second of finished video. 285 models · per second Picked by lowest price in the band | under $0.06 · 100 models bytedance/seedance-1.5-pro, without audio-480p (service ends 2026-11-11) $0.0088 kie.ai | ||
| Text to speech Voice output, priced by 1,000 characters of input text. 7 models · per 1k characters Picked by lowest price in the band | |||
| Music generation One prompt in, one finished track out. 8 models · per track Picked by lowest price in the band |
Read it like this
Each column is a budget, not a quality tier. The Standard column carries the yellow rule because it is where most production work lands, not because those models are better.
How a pick is made
Same rule every day- Take every live-fetched, non-duplicate model whose billing unit is directly comparable for that job. Different units are never mixed.
- Drop every $0 row. A free tier is a trial, not a price.
- Split what is left into three fixed price bands. The boundaries are printed in the table and do not move with the data.
- Inside a band, pick by the one capability figure the dataset carries. For text that is the largest context window, tie broken by the lower price. For image, video and voice, where no capability figure is published, it is the lowest price.
What this is not
It is not a quality ranking. APITariff measures prices, and a price cannot tell you which model writes better code. Where a pick is made on price alone, the table says so.
Own benchmark scores are being collected now. When they publish, step 4 changes from lowest price to best score per dollar, in this same table, and the change gets a dated entry in the ledger.
Full methodology · Cheapest LLM by output price, all 482 text models ranked
Band boundaries
Fixed, published, unchanged by the data| Job | Unit | Budget | Standard | Premium | Models |
|---|---|---|---|---|---|
| Text generationper 1M output tokens | per 1M output tokens | under $1 | $1 to $10 | $10 and up | 568 |
| Long-context textper 1M output tokens | per 1M output tokens | under $1 | $1 to $10 | $10 and up | 323 |
| Image generationper image | per image | under $0.03 | $0.03 to $0.10 | $0.10 and up | 278 |
| Video generationper second | per second | under $0.06 | $0.06 to $0.20 | $0.20 and up | 285 |
| Text to speechper 1k characters | per 1k characters | under $0.03 | $0.03 to $0.10 | $0.10 and up | 7 |
| Music generationper track | per track | under $0.05 | $0.05 to $0.15 | $0.15 and up | 8 |