AIインサイト/Global Frontier LLM API Pricing × Token Structure (June 2026)
業界インサイト2026-06·読了目安:6 min

Global Frontier LLM API Pricing × Token Structure (June 2026)

One ruler for 20 high-traffic frontier LLMs: Chinese models average ~6–9x cheaper than US ones — yet input dwarfs output by tens of times and cache covers 50–94% of input, so the bill is driven by input & cache pricing, not the headline output price. Full price × token-structure table and a cost-formula breakdown included.

We put 20 frontier large language models — those with real, at-scale API traffic, image/video models excluded — onto one ruler, measured on two axes: price (USD per 1M tokens) and token structure (the in : cache : out : reasoning mix of real usage, normalized to output = 1).

The headline gap is regional. On average a US model charges about 8.9× more per output token and 5.9× more per input token than a Chinese one. But output is the smallest slice of real workloads — so the real bill is driven by input and cache pricing, not the sticker output price.

1 · Two price bands

🇺🇸 US
Avg input
$2.93
Avg output
$15.43
Avg cache
$0.3
🇨🇳 China
Avg input
$0.5
Avg output
$1.73
Avg cache
$0.08

US flagships (Claude Opus, GPT-5.5) sit at $5 in / $25–30 out. Chinese models cluster an order of magnitude below — the cheapest, Tencent Hy3, is $0.063 in / $0.21 out, and DeepSeek V4 Flash undercuts it on output at $0.18.

2 · The full table

ModelInputOutputCacheIn : Cache : Out : Think
US · avg in $2.93 / out $15.43 / cache $0.30
Claude Opus 4.7
Claude · 2026-04 · 1M
$5$25$0.5114.6 : 94 : 1 : 0.05
Claude Opus 4.8
Claude · 2026-05 · 1M
$5$25$0.586.2 : 68.3 : 1 : 0.06
GPT-5.5
GPT · 2026-04 · 1.05M
$5$30$0.567.1 : 55.9 : 1 : 0.51
GPT-5.4
GPT · 2026-03 · 1.05M
$2.5$15$0.2538.3 : 22.8 : 1 : 0.23
Gemini 3.1 Flash Lite
Gemini · 2026-05 · 1M
$0.25$1.5$0.02511.6 : 3.5 : 1 : 0.31
Gemini 3.5 Flash
Gemini · 2026-05 · 1M
$1.5$9$0.1517.1 : 10.4 : 1 : 0.7
Grok 4.3
Grok · 2026-04 · 1M
$1.25$2.5$0.28.8 : 4.4 : 1 : 0.68
China · avg in $0.50 / out $1.73 / cache $0.08
DeepSeek V4 Flash
DeepSeek · 2026-04 · 1M
$0.09$0.18$0.0223.5 : 13.4 : 1 : 0.52
DeepSeek V4 Pro
DeepSeek · 2026-04 · 1M
$0.44$0.87$0.003623.9 : 19.5 : 1 : 0.7
MiMo-V2.5
小米 MiMo · 2026-04 · 1M
$0.14$0.28$0.0028111.4 : 104.2 : 1 : 0.36
MiMo-V2.5-Pro
小米 MiMo · 2026-04 · 1M
$0.44$0.87$0.003682.6 : 73.2 : 1 : 0.38
MiniMax M3
MiniMax · 2026-05 · 1M
$0.3$1.2$0.0653 : 44.6 : 1 : 0.53
Hy3 preview
腾讯混元 · 2026-04 · 262K
$0.063$0.21$0.02160 : 53.9 : 1 : 0.37
GLM 5.2
GLM · 2026-06 · 1M
$0.98$3.08$0.1861.2 : 48.3 : 1 : 0.56
GLM 5.1
GLM · 2026-04 · 203K
$0.98$3.08$0.1852.9 : 45.2 : 1 : 0.62
Step 3.7 Flash
阶跃星辰 · 2026-05 · 256K
$0.2$1.15$0.0450.2 : 35.3 : 1 : 0.02
Kimi K2.6
Kimi · 2026-04 · 262K
$0.66$3.41$0.1431.3 : 23.7 : 1 : 0.78
Kimi K2.7 Code
Kimi · 2026-06 · 262K
$0.61$3.07$0.1358.2 : 49.3 : 1 : 0.64
Qwen3.7 Plus
通义千问 · 2026-06 · 1M
$0.32$1.28$0.06420.5 : 15.1 : 1 : 0.71
Qwen3.7 Max
通义千问 · 2026-05 · 1M
$1.25$3.75$0.2528.1 : 21.1 : 1 : 0.69
Price: USD / 1M tokens; token structure normalized to output = 1.

3 · Token structure: input dwarfs output

In real usage every output token rides on tens of input tokens. MiMo-V2.5 runs 111 : 104 : 1 — 111 input tokens per output token, of which 104 are cache hits (94%). Even the leanest, Grok 4.3, is 8.8 : 4.4 : 1. Reasoning tokens stay small (mostly under 0.7 of output); Claude and Step barely meter them.

High cache ratios (50–94% of input) mean heavy prompt reuse — system prompts, documents, codebases sent again and again. Cache pricing (typically ~1/10 of input) is therefore the hidden lever on cost.

4 · What you actually pay

Total cost = (input − cache) × input price + cache × cache price + output (incl. reasoning) × output price.

Because cached input bills at the discounted cache price, and input outweighs output by tens of times, the real bill is dominated by how much of your input hits cache — not by the headline output price. A pricier model with aggressive cache discounts can end up cheaper in production.

Method: API list prices as of June 2026 (USD / 1M tokens); token structure from real-usage statistics of mainstream models, normalized to output = 1; 20-model sample with at-scale traffic, image/video models excluded.

関連記事

Updated monthly
AI Traffic & App Statistics

AI chatbot market share, fastest-growing AI tools, categories and countries, plus AI app downloads and in-app purchase revenue.

2026-09
AI Gateways: Crowded, Leader Gains (Sep 2026)

74 new entrants in twelve months, traffic up 69% against a falling market — and leader OpenRouter gained share rather than losing it, from 37.4% to 49.3%. The breakdown: 21.6% of the growth was carried in by new supply, incumbent demand grew only 32.7%, and China supplies 19.3% of visits — 4.8x its share of the wider market. Full Top 50 included.

2026-11
Anthropic IPO: Claude Beyond the Chat Window

Thirteen months of Claude web and app data, six months of Chrome extension growth, Artificial Analysis benchmarks and prospectus disclosures show how Anthropic reaches consumers and enterprises.

2026-07
WAIC 2026 Shanghai: AI in Four Niche Verticals (July 2026)

Across ~70 booths in the H4 industry-application street, the busiest theme was niche-vertical AI agents — beyond content creation and audio/video generation, there were plenty of products for smart office, research, user-interview and legal work.

← AIインサイトに戻る