
At the platform's effective price, the cheapest text model right now is gpt-6-luna ($0.0045 per million input tokens), the cheapest image model is krea-2-turbo ($0.0038 per image), and the cheapest video model is hailuo-2_3-fast ($0.0017 per second). Data as of November 16, 2026.
Different models bill in different units — text by token, images by image, video by second, audio by minute. Dropping "$0.0038 per image" and "$0.0045 per million tokens" into one ranking is comparing apples to oranges. So I split this into three lists, each comparing models that bill in the same unit.
"Cheap" here means the effective price — the official price multiplied by the seller's discount — not the list price. A $100 model at a 0.5 discount costs $0.5; a $10 model at a 1.0 discount costs $0.1. Judge by what you actually pay, not the number on the vendor's pricing page.
| Model | Vendor | Official ($/M input) | Effective | Discount |
|---|---|---|---|---|
| gpt-6-luna | [OI] | $0.1 | $0.0045 | 0.45 |
| claude-haiku-5-5 | Anthropic | $0.1 | $0.005 | 0.5 |
| muse-spark-1.2-contributor | Meta | $0.1 | $0.005 | 0.5 |
The cheapest text models cluster in the $0.1-per-million entry tier. Note this is the input price only — output usually costs about five times input, so estimate both before you commit.
| Model | Official ($/image) | Effective | Discount |
|---|---|---|---|
| krea-2-turbo | $0.015 | $0.0038 | 2.5 |
| concept-art | $0.028 | $0.007 | 2.5 |
| graphic-design | $0.028 | $0.007 | 2.5 |
Image models are simpler than text: a flat price per image, no input/output split. The spread is what matters — the cheapest and most expensive differ by tens of times, so decide whether you need "good enough" or "high quality" before you pick.
| Model | Official ($/second) | Effective | Discount |
|---|---|---|---|
| hailuo-2_3-fast | $0.0336 | $0.0017 | 0.5 |
| hailuo-2_3 | $0.0498 | $0.0025 | 0.5 |
| kling-2.5-turbo-standard | $0.0448 | $0.0112 | 2.5 |
Video bills per second, so a ten-second clip on the cheapest model costs about $0.017. But "cheap" hides resolution and duration tiers — the low tier and high tier differ a lot, so read the tier table before you buy.
Of the 107 models with at least one quote, 77 sit in the 0.1–0.5 band, 12 sit at the 5.0 ceiling (the hard cap — sellers may not quote above 50% of list price), and only two sit in the middle 3–5 band.
That's not random. 5.0 is the ceiling and 0.5 is the lowest most sellers accept. The middle is neither cheap enough to win a request nor profitable enough to bother with, so nobody parks there.
The distribution is also a signal: a model stuck at 5.0 means supply is thin; one below 0.5 means several sellers are competing for you.
Run 1M input tokens through gpt-6-luna: $0.1 at list price, $0.0045 effective — 95.5% saved. A month of 10M input plus 1M output costs about $10 officially and around 50 cents on this model.
krea-2-turbo for images: $0.015 official, $0.0038 effective, 75% saved. A hundred material images cost $1.5 officially versus $0.38 here.
hailuo-2_3-fast for video: $0.0336 per second official, $0.0017 effective, 95% saved.
None of this is a limited-time promo. Sellers list idle quota at their own prices — that's the whole reason it undercuts direct official access.
Three honest caveats.
First, cheap and reliable are different things. Many low-price listings carry a "no calls yet" flag — the seller posted capacity nobody has actually exercised. For production, filter for certified sellers, not just the lowest number.
Second, cheap models are limited. A $0.1-tier model like gpt-6-luna handles summaries, classification and tagging fine, but not complex reasoning.
Third, prices move. Sellers post and withdraw asks constantly; today's cheapest may be gone tomorrow. Go by the live price at order time.
Three things you can act on today: use the cheap text models for batch work (classification, summaries, tagging) at near-zero cost; use krea-2-turbo for material images at under $0.004 each; use hailuo-2_3-fast for short test clips at $0.017 per ten seconds.
Check the live price table for current numbers instead of memorising these, and browse model docs for each model's billing unit, tier prices and runnable examples.
GPT-6.2 cuts list prices 20%, yet platform discounts run deeper: gpt-6-astra trades at under 1% of list. What this week's AI price war means for buyer
Prompt caching stores a repeated prompt prefix and bills cache reads at 5-12% of normal input. Real prices, hit conditions, and how to cut your API bi
Twelve Claude versions are on sale. The newest opus-5-5 costs $8.91 a month; the oldest opus-4-6 costs $49.50. Full table and a pick for each workload
gpt-5.6-terra and deepseek-v4.1-flash win on price and reliability; gemini-3.8-flash is fastest. Effective prices, measured speed, and a three-tier pi
Point your base URL at the platform and one key covers all 138 on-sale models across 51 vendors, in both OpenAI and Anthropic request formats. Copy-pa