For developers

API pricing: the real bill, per million tokens.

Consumer plans are flat-rate; the API is metered. Rates below are per million tokens (≈750k words), input / output, sorted cheap to expensive. Verified July 2026 — these change often.

ProviderModelInput /1MOutput /1MContext
OpenAIGPT-5.4 nano
Fastest, cheapest
$0.05$0.40272K
DeepSeekV4-Flash
Fast, cheap reasoning
$0.14$0.281M
GoogleGemini 3.1 Flash-Lite
Cheap, fast, multimodal
$0.25$1.501M
DeepSeekV4-Pro
Top reasoning, cheap
$0.43$0.871M
AnthropicClaude Haiku 4.5
Fast, near-Sonnet quality
$1.00$5.00200K
xAIGrok 4.5
Opus-class flagship
$2.00$6.00500K
GoogleGemini 3.1 Pro
Frontier multimodal
$2.00$12.001M
OpenAIGPT-5.6 Terra
Balanced flagship tier
$2.50$15.001M
AnthropicClaude Sonnet 5
New Sonnet flagship
$3.00$15.001M
AnthropicClaude Opus 4.8
Opus flagship
$5.00$25.001M
OpenAIGPT-5.6 Sol
New flagship (Jul 2026)
$5.00$30.001M
AnthropicClaude Fable 5
Most powerful Claude
$10.00$50.001M
OpenAIGPT-5.5 pro
Most capable overall
$30.00$180.001M

Worked example: a 2,500-word essay (≈6,700 tokens, half in/half out)

GPT-5.4 nano ≈ $0.002 · DeepSeek V4-Flash ≈ $0.002 · Claude Haiku 4.5 ≈ $0.02 · Claude Sonnet 5 ≈ $0.06 · Claude Opus 4.8 ≈ $0.10 · GPT-5.5 pro ≈ $0.45. The cheapest model is usually 95% as good for simple tasks — start cheap, escalate on failure.

Building for a team instead of an app? Multi-model access without writing code is what workspaces like AI bundles are for — and its Enterprise tier supports bring-your-own API keys if you want metered billing with a managed UI.

FAQ

API questions.

How do API prices work?

You pay per million tokens — roughly 750,000 words. Output tokens typically cost 3–6× input tokens. A 2,500-word translation costs anywhere from under a cent (mini models) to ~$0.45 (top reasoning models).

Which API is the best value in 2026?

For cheap capable output: DeepSeek V4-Flash ($0.14/$0.28) and GPT-5.4 nano ($0.05/$0.40). For quality-per-dollar in the mid tier: Claude Haiku 4.5 and Gemini 3.1 Flash-Lite. Start cheap and upgrade only when output quality actually fails.

Do I need the API or a subscription?

If you chat with AI, subscribe. The API is for developers building products — and for chat, subscriptions are almost always cheaper than paying per token at the flagship rates.