What is the cheapest frontier AI API in July 2026?

DeepSeek V4 Flash is the cheapest actively-supported frontier-class API at $0.14 per million input tokens and $0.28 output (off-peak) — about $0.017 on a 30K-in/5K-out task. GPT-5.6 Luna ($1/$6), Gemini 3.6 Flash ($1.50/$7.50), and Grok 4.5 ($2/$6) are the cheapest closed options.

How do I calculate cost per task for an AI model?

Use: cost = (input_tokens/1,000,000 × input_price) + (output_tokens/1,000,000 × output_price). For a 30K-input/5K-output task on Gemini 3.6 Flash ($1.50/$7.50): (0.03 × 1.50) + (0.005 × 7.50) = $0.045 + $0.0375 = $0.0825.

Is the cheapest model good enough for production?

For classification, extraction, summarization, and routine coding — yes. DeepSeek V4 Flash and Gemini 3.6 Flash are frontier-class at those tasks. For the hardest reasoning and long-horizon agents, escalate to a frontier model like Claude Opus 5 or GPT-5.6 Sol only when needed.

Quick Answer

Cheapest Frontier AI API July 2026: Cost Per Task

Published: July 28, 2026

The Short Answer

DeepSeek V4 Flash is the cheapest frontier-class API in late July 2026 at $0.14/$0.28 per MTok (off-peak) — roughly $0.017 per typical task. Among closed models, GPT-5.6 Luna ($1/$6), Gemini 3.6 Flash ($1.50/$7.50), and Grok 4.5 ($2/$6) are the value leaders.

Ranked by Cost Per Task

Task = 30K input + 5K output tokens. Formula: 30×(in/1000) + 5×(out/1000) in cents.

Model	In/Out ($/MTok)	Cost per task	Type
DeepSeek V4 Flash	$0.14 / $0.28	~$0.0056*	open
DeepSeek V4 Pro	$0.435 / $0.87	~$0.017*	open
GPT-5.6 Luna	$1 / $6	~$0.06	closed
Gemini 3.6 Flash	$1.50 / $7.50	~$0.083	closed
Grok 4.5	$2 / $6	~$0.09	closed
Gemini 3.5 Flash-Lite	$0.30 / $2.50	~$0.021	closed

*DeepSeek applies a 2x surcharge during peak UTC windows (~01–04 & 06–10 UTC), so budget double at peak.

The Cheapest by Category

Absolute cheapest (open) → DeepSeek V4 Flash. MIT-licensed, single-80GB-GPU self-host option, and the lowest per-token API rate of any frontier-class model.
Cheapest closed → GPT-5.6 Luna at $1/$6, or Gemini 3.5 Flash-Lite at $0.30/$2.50 for very high-volume, simpler tasks.
Cheapest with a huge context → DeepSeek V4 (1M tokens) or Gemini 3.6 Flash, both cheap enough to feed large documents without pain.

Watch the Fine Print

Peak pricing: DeepSeek’s headline rate is off-peak only; the peak 2x can flip your ranking if your traffic clusters in UTC mornings.
Retirements: Gemini 2.5 Flash-Lite is cheaper still but retires October 16, 2026 — don’t build new integrations on it.
Token counts vary: newer tokenizers can inflate effective token counts. Benchmark on your prompts, not the sticker price.

The Real Lever: Routing

Even the cheapest single model isn’t the cheapest architecture. Route: send the easy 80% of requests to DeepSeek V4 Flash or Gemini 3.6 Flash, and escalate only hard requests to a frontier model. Most teams cut spend 5-10x versus running everything on a flagship.

Sources

DeepSeek — V4 pricing: deepseek.ai/pricing
Google — Gemini API pricing: ai.google.dev/pricing
VentureBeat — Gemini 3.6 Flash cuts token costs: venturebeat.com
CloudZero — DeepSeek pricing: cloudzero.com/blog/deepseek-pricing