How This Comparison Works
Comparing AI API pricing means comparing three different billing units at once: text models bill per token, image models per image (or per output token), and video models per second or per minute. This guide normalizes each category, lists verified rates from the major providers, and shows where an aggregator fits. All provider figures are standard rates from official pricing pages as of mid-2026 — verify before budgeting, as they change often.
A quick note on categories: OpenAI, Google, and Anthropic sell first-party models. LinkModel is an aggregation gateway — it resells models from these and other providers behind one API at up to 30% below official rates, with a focus on image and video generation. So the honest framing isn't "LinkModel vs OpenAI" as rival labs, but "buy direct vs buy aggregated."
Text Models (per 1M tokens, input / output)
| Model | Provider | Input | Output |
|---|---|---|---|
| GPT-5.5 | OpenAI | $5.00 | $30.00 |
| GPT-5.4 | OpenAI | $2.50 | $15.00 |
| GPT-5.4 Nano | OpenAI | $0.20 | — |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 |
| Gemini 3.1 Pro | ~$2.00 | ~$12.00 | |
| Gemini 2.5 Flash | $0.30 | $2.50 | |
| DeepSeek V4 Flash | DeepSeek | $0.14 | $0.28 |
Takeaways: OpenAI and Anthropic anchor the premium tier ($5/$25–30); Google's Pro sits a notch below; DeepSeek's V4 Flash is dramatically cheaper for output-heavy work. Every provider offers ~50% Batch discounts and prompt caching (up to ~90% off cached input) that reshape effective cost. Details per vendor in the Claude Opus pricing guide.
Image Models
| Model | Provider | Cost per image |
|---|---|---|
| GPT Image 2 | OpenAI | ~$0.05 (med) / ~$0.21 (high) |
| Nano Banana Pro | $0.134 (1K/2K) / $0.24 (4K) | |
| Nano Banana 2 | Lower (Flash tier) | |
| Seedream | ByteDance | ~$0.03–0.04 |
Premium image models cluster around $0.05–0.24; bulk/open models drop an order of magnitude. Full breakdowns: GPT Image 2 pricing, Gemini image pricing, and the best image APIs roundup.
Video Models (normalized $/min of 1080p)
| Model | Provider | ~$/min | Note |
|---|---|---|---|
| Seedance 2.0 | ByteDance | ~$9 | Arena leader (with audio) |
| Kling V3 (Pro) | Kuaishou | ~$20 | Native 4K |
| Veo 3.1 | ~$24 | 4K, native audio | |
| Sora 2 | OpenAI | ~$18–30 | API sunsets Sep 2026 |
Full field in best AI video generation APIs.
Where the Aggregator Fits
Buying direct gives you the newest models first and each provider's Batch discount, but you juggle multiple keys, bills, SDKs, and — on per-GPU-second platforms — variable costs. An aggregator trades a little of that immediacy for consolidation: one key, one bill, one request shape, and a structural discount. LinkModel prices generation models up to 30% below official with fixed per-request pricing (the price shows before each call), a 99.95% SLA, and zero data retention by default. For a $10k/month generation bill, a 25–30% markdown is $2,500–$3,000/month before any optimization.
The honest caveat: an aggregator's model coverage and exact per-model discount vary and change, so treat "up to 30%" as a ceiling and confirm live figures on the pricing page and model catalog. The structural comparison against a per-GPU-second platform is in LinkModel vs fal.ai.
Practical Decision
- Latest frontier text model, run in batches → buy direct from the provider.
- Cheapest text for high volume → DeepSeek V4 Flash (mind data residency).
- Commercial image/video generation at a predictable discount → aggregate through LinkModel.
- Open-weight / self-host → serverless GPU platforms.
The single biggest cost lever isn't the provider — it's routing simple work to cheap models and reserving premium ones for hard tasks. See how to reduce AI API costs and, for the generation-cost floor, the cheapest AI API guide.
Start on LinkModel with a $1 credit and compare your real per-call cost against buying direct.