Cheapest AI API August 2026: Price Per Token Ranked

The Short Answer

The cheapest AI API in August 2026 is DeepSeek V4-Flash ($0.14/$0.28) at the frontier-adjacent tier. For lighter tasks, Amazon Nova Micro, Qwen 3.7 Flash, and Gemini 3.5 Flash-Lite go even lower.

The Ranking (per million tokens)

Rank	Model	Input	Output	Tier
1	Llama 3.1 8B Instruct	$0.02	~$0.03	Small/cheapest overall
2	Qwen 3.7 Flash	$0.03	$0.13	Light coding
3	Amazon Nova Micro	$0.035	$0.14	Light/fast
4	Amazon Nova Lite	$0.06	$0.24	Light
5	DeepSeek V4-Flash	$0.14	$0.28	Frontier-adjacent
6	Gemini 3.5 Flash-Lite	$0.30	$2.50	Fast hosted
—	Gemini 3.6 Flash	$1.50	$7.50	Value workhorse

How To Read This

• Best value at the frontier → DeepSeek V4-Flash. The “price floor of the frontier-adjacent market” — roughly 10x-90x cheaper than Claude Opus 5 and GPT-5.6 Sol. Cache-hit input is just $0.0028/MTok, so cache-heavy agent loops cost almost nothing.
• Absolute lowest for light tasks → Llama 3.1 8B, Qwen 3.7 Flash, and Amazon Nova Micro — great for classification, extraction, and short responses, but not frontier reasoning.
• Cheap hosted convenience → Gemini 3.5 Flash-Lite ($0.30/$2.50) and Gemini 3.6 Flash ($1.50/$7.50), which uses up to 65% fewer output tokens.

Watch Outs

• DeepSeek peak pricing: a 2x peak-hour surcharge (Beijing-time windows) has been announced but has no effective date as of early August 2026 — standard rates still apply. The deepseek-chat / deepseek-reasoner names were retired July 24, 2026.
• Cheap ≠ capable: Nova Micro and Llama 8B are small models. For hard multi-step tasks, DeepSeek V4-Flash or a flagship is worth the higher price.

Verdict

• Cheapest frontier-adjacent → DeepSeek V4-Flash
• Cheapest overall → Llama 3.1 8B / Qwen 3.7 Flash
• Cheapest hosted mid-tier → Gemini 3.5 Flash-Lite

Sources

• DeepSeek — pricing: deepseek.ai/pricing
• AWS — Amazon Nova: aws.amazon.com/ai/generative-ai/nova
• Google — Gemini release notes: gemini.google/release-notes