AI Price Index: September 2026

Rebuilt on 2026-11-04 from the models.dev history · frozen, not revised afterwards · full index and methodology

In September 2026, the median AI model cost $1.50 per 1M tokens on labs' own APIs, +20% on the previous month. 14 models added, 10 retired, 2 repriced.

index123.7January 2026 = 100
median price$1.50+20% vs last month
budget / premium$0.218 / $8.8010th / 90th percentile
tracked177models from 12 labs

// list price changes

modellabinput $/1Moutput $/1Mblended
Qwen3.7 PlusAlibaba (Qwen)$0.50 → $0.40$3.00 → $1.60−37.8%
GLM-5.3-FlashZ.AI (Zhipu)$0.075 → $0.15$0.25 → $0.50+99.9%

// added this month (14)

modellabinput $/1Moutput $/1Mblended
Claude Fable 5.1Anthropic$10.00$50.00$20.00
GPT-6 AstraOpenAI$10.00$50.00$20.00
Claude Opus 5.5Anthropic$4.00$20.00$8.00
Claude Sonnet 5.5Anthropic$2.00$10.00$4.00
GPT-6 SolOpenAI$2.00$10.00$4.00
GPT-6.1 SolOpenAI$2.00$10.00$4.00
Grok 4.7xAI$2.00$6.00$3.00
Muse Spark 1.3Meta$1.25$4.25$2.00
Gemini 3.8 FlashGoogle$0.75$3.75$1.50
DeepSeek V4.1 FlashDeepSeek$0.15$0.60$0.2625
Qwen3.8 Omni FlashAlibaba (Qwen)$0.15$0.47$0.23
GPT-6 LunaOpenAI$0.10$0.50$0.20
Qwen3.5 FlashAlibaba (Qwen)$0.10$0.40$0.175
Qwen3.7 FlashAlibaba (Qwen)$0.03$0.13$0.055

New in the catalog this month, which can include renamed models or older models added late.

// retired this month (10)

DeepSeek V4 Flash ($0.175) · DeepSeek V4 Flash Vision Exp ($0.175) · Gemini Robotics-ER 1.6 Preview ($2.00) · Kimi K2 0711 ($1.075) · Kimi K2 0905 ($1.075) · Kimi K2 Thinking ($1.075) · Kimi K2 Thinking Turbo ($2.8625) · Kimi K2 Turbo ($4.30) · Kimi K2.5 ($1.20) · Magistral Small ($0.75)

// median price by lab, where it moved

labAugust 2026September 2026change
DeepSeek$0.175$0.4032+130.4%
OpenAI$4.50$4.375−2.8%
Mistral$0.725$0.70−3.4%
Alibaba (Qwen)$0.90$0.85−5.6%

// cheapest model by capability

capabilitymodelblended
tool calling, 128K+ contextQwen3.7 Flash$0.055
reasoning and tool callingQwen3.7 Flash$0.055
1M+ contextQwen3.7 Flash$0.055
image inputQwen3.7 Flash$0.055

← August 2026October 2026 →full indexevery model (json)