No subscription tiers, no surge pricing, no margin layered on top. We pass through provider rates as published, debit your wallet in cents, and show you the same numbers our proxy uses. Last verified 2026-04-30.
"Coding minutes" assumes ~3,000 tokens/min wall-clock active throughput, the rough average for an agentic coding session that mixes thinking, writing, and reading code. A single 30-second video ad pays out roughly 3¢ on current AppLixir CPMs — the numbers above show what that one ad funds, per lane.
AdToken routes Lane 1 through Vercel AI Gateway, which has a commercial reseller agreement with Anthropic. No BYO key required. Choose a faster/cheaper model with --model=anthropic/claude-haiku-4-5 or step up to --model=anthropic/claude-opus-4-7.
| Model | Input $/M | Output $/M | Blended $/M | Tokens / 3¢ | ≈ Coding / 3¢ |
|---|---|---|---|---|---|
Claude Sonnet 4.6anthropic:sonnet-4-6 |
$3.00 | $15.00 | $9.00 | 3,333 | 1.1 min |
Claude Haiku 4.5anthropic:haiku-4-5 |
$0.80 | $4.00 | $2.40 | 12,500 | 4.2 min |
Claude Opus 4.7anthropic:opus-4-7 |
$15.00 | $75.00 | $45.00 | 667 | 0.2 min |
Same shape as Lane 1 — Vercel AI Gateway resells OpenAI on AdToken's account. No BYO key required. o-series reasoning models bill hidden reasoning tokens at the output rate, so real spend on o3/o4-mini can be 3–10× the visible response on coding tasks.
| Model | Input $/M | Output $/M | Blended $/M | Tokens / 3¢ | ≈ Coding / 3¢ |
|---|---|---|---|---|---|
GPT-5openai:gpt-5 |
$1.25 | $10.00 | $5.63 | 5,333 | 1.8 min |
GPT-5 miniopenai:gpt-5-mini |
$0.25 | $2.00 | $1.13 | 26,667 | 8.9 min |
GPT-5.5openai:gpt-5.5 |
$5.00 | $30.00 | $17.50 | 1,714 | 0.6 min |
OpenAI o3openai:o3 |
$2.00 | $8.00 | $5.00 | 6,000 | 2.4 min |
OpenAI o4-miniopenai:o4-mini |
$1.10 | $4.40 | $2.75 | 10,909 | 4.4 min |
Open-source models routed through Vercel AI Gateway (primary) or OpenRouter (fallback). 10–50× cheaper than Lane 1/2 on a $/M basis — one ad funds dramatically more coding here, which is why this is the production lane for the closed ad-funded loop.
| Model | Input $/M | Output $/M | Blended $/M | Tokens / 3¢ | ≈ Coding / 3¢ |
|---|---|---|---|---|---|
Llama 4 405B (Meta, US)openrouter:llama-4-405b |
$0.40 | $1.20 | $0.80 | 37,500 | 12 min |
GPT-OSS 120B (OpenAI, US)openrouter:gpt-oss-120b |
$0.15 | $0.60 | $0.38 | 80,000 | 27 min |
GPT-OSS 20B (OpenAI, US)openrouter:gpt-oss-20b |
$0.05 | $0.20 | $0.13 | 240,000 | 80 min |
DeepSeek V3 (China)openrouter:deepseek-v3 |
$0.27 | $1.10 | $0.69 | 43,796 | 15 min |
Kimi K2 (Moonshot, China)openrouter:kimi-k2 |
$0.15 | $2.50 | $1.32 | 22,642 | 7.5 min |
Qwen 3 32B (Alibaba, China)openrouter:qwen-3-32b |
$0.10 | $0.30 | $0.20 | 150,000 | 50 min |
Mistral Large (Mistral, France)openrouter:mistral-large |
$2.00 | $6.00 | $4.00 | 7,500 | 2.5 min |
DeepSeek V4 Pro (via Vercel AI Gateway)vercel:deepseek-v4-pro |
$0.43 | $0.87 | $0.65 | 46,154 | 15 min |
DeepSeek V4 Flash (via Vercel AI Gateway)vercel:deepseek-v4-flash |
$0.14 | $0.28 | $0.21 | 142,857 | 48 min |
Qwen 3.6 27B (Alibaba, via Vercel)vercel:qwen-3.6-27b |
$0.60 | $3.60 | $2.10 | 14,286 | 4.8 min |
Qwen 3.6 Plus (Alibaba, via Vercel)vercel:qwen-3.6-plus |
$0.50 | $3.00 | $1.75 | 17,143 | 5.7 min |
Qwen 3.6 Max Preview (Alibaba, via Vercel)vercel:qwen-3.6-max-preview |
$1.30 | $7.80 | $4.55 | 6,593 | 2.2 min |
Kimi K2.6 (Moonshot, via Vercel)vercel:kimi-k2.6 |
$0.95 | $4.00 | $2.48 | 12,121 | 4.0 min |
Every proxy call computes spend the same way:
cents = round( (input_tokens × input_per_m + output_tokens × output_per_m) / 10_000 )
The PRICING table is the only source of those per_m numbers. Whatever you see on this page is exactly what your wallet gets debited in /credits. There is no hidden margin layer between this page and the database write.
Aggregator margin (Vercel AI Gateway is 0% as of 2026-04-30; OpenRouter charges 5–10% on the fallback path) is absorbed by AdToken's operator pot — your wallet only ever sees the published provider rate.