钽TANDIU钽铥
TOKEN PRICE INDEX

LLM Price Index

Official LLM API prices as a trackable index: US vs Chinese price trends, the cheapest model at a given benchmark score, and GPU rental prices. Updated daily; data downloadable.

Data updated: 2026-10-06
Chinese models114.630-day -0.9% · median $1.54 · 26 models
US models98.730-day +0.0% · median $4.00 · 39 models
All models104.930-day -0.4% · median $2.00 · 66 models
9710210711211707-1308-0308-2409-1510-06115Chinese models 2026-07-13: 100Chinese models 2026-07-14: 100Chinese models 2026-07-15: 100Chinese models 2026-07-16: 100Chinese models 2026-07-17: 100Chinese models 2026-07-18: 100Chinese models 2026-07-19: 100Chinese models 2026-07-20: 100Chinese models 2026-07-21: 100Chinese models 2026-07-22: 100Chinese models 2026-07-23: 100Chinese models 2026-07-24: 100Chinese models 2026-07-25: 100Chinese models 2026-07-26: 100Chinese models 2026-07-27: 100Chinese models 2026-07-28: 100Chinese models 2026-07-29: 100Chinese models 2026-07-30: 100Chinese models 2026-07-31: 100Chinese models 2026-08-01: 100Chinese models 2026-08-02: 100Chinese models 2026-08-03: 100Chinese models 2026-08-04: 100Chinese models 2026-08-05: 100Chinese models 2026-08-06: 100Chinese models 2026-08-07: 100Chinese models 2026-08-08: 100Chinese models 2026-08-09: 100Chinese models 2026-08-10: 100Chinese models 2026-08-11: 100Chinese models 2026-08-12: 100Chinese models 2026-08-13: 100Chinese models 2026-08-14: 100Chinese models 2026-08-15: 100Chinese models 2026-08-16: 100Chinese models 2026-08-17: 100Chinese models 2026-08-18: 100Chinese models 2026-08-19: 100Chinese models 2026-08-20: 100Chinese models 2026-08-21: 116Chinese models 2026-08-22: 116Chinese models 2026-08-23: 116Chinese models 2026-08-24: 116Chinese models 2026-08-25: 116Chinese models 2026-08-26: 116Chinese models 2026-08-27: 116Chinese models 2026-08-28: 116Chinese models 2026-08-29: 116Chinese models 2026-08-30: 116Chinese models 2026-08-31: 116Chinese models 2026-09-01: 116Chinese models 2026-09-02: 116Chinese models 2026-09-03: 116Chinese models 2026-09-04: 116Chinese models 2026-09-05: 116Chinese models 2026-09-06: 116Chinese models 2026-09-07: 116Chinese models 2026-09-08: 116Chinese models 2026-09-09: 116Chinese models 2026-09-10: 116Chinese models 2026-09-11: 116Chinese models 2026-09-12: 116Chinese models 2026-09-13: 114Chinese models 2026-09-14: 114Chinese models 2026-09-15: 114Chinese models 2026-09-16: 114Chinese models 2026-09-17: 114Chinese models 2026-09-18: 114Chinese models 2026-09-19: 114Chinese models 2026-09-20: 114Chinese models 2026-09-21: 114Chinese models 2026-09-22: 115Chinese models 2026-09-23: 115Chinese models 2026-09-24: 115Chinese models 2026-09-25: 115Chinese models 2026-09-26: 115Chinese models 2026-09-27: 115Chinese models 2026-09-28: 115Chinese models 2026-09-29: 115Chinese models 2026-09-30: 115Chinese models 2026-10-01: 115Chinese models 2026-10-02: 115Chinese models 2026-10-03: 115Chinese models 2026-10-04: 115Chinese models 2026-10-05: 115Chinese models 2026-10-06: 11599US models 2026-07-13: 100US models 2026-07-14: 100US models 2026-07-15: 100US models 2026-07-16: 100US models 2026-07-17: 100US models 2026-07-18: 100US models 2026-07-19: 100US models 2026-07-20: 100US models 2026-07-21: 100US models 2026-07-22: 100US models 2026-07-23: 100US models 2026-07-24: 100US models 2026-07-25: 100US models 2026-07-26: 100US models 2026-07-27: 100US models 2026-07-28: 100US models 2026-07-29: 100US models 2026-07-30: 100US models 2026-07-31: 100US models 2026-08-01: 100US models 2026-08-02: 100US models 2026-08-03: 100US models 2026-08-04: 100US models 2026-08-05: 100US models 2026-08-06: 100US models 2026-08-07: 100US models 2026-08-08: 100US models 2026-08-09: 100US models 2026-08-10: 100US models 2026-08-11: 100US models 2026-08-12: 100US models 2026-08-13: 100US models 2026-08-14: 100US models 2026-08-15: 100US models 2026-08-16: 100US models 2026-08-17: 100US models 2026-08-18: 100US models 2026-08-19: 100US models 2026-08-20: 100US models 2026-08-21: 100US models 2026-08-22: 100US models 2026-08-23: 99US models 2026-08-24: 99US models 2026-08-25: 99US models 2026-08-26: 99US models 2026-08-27: 99US models 2026-08-28: 99US models 2026-08-29: 99US models 2026-08-30: 99US models 2026-08-31: 99US models 2026-09-01: 99US models 2026-09-02: 99US models 2026-09-03: 99US models 2026-09-04: 99US models 2026-09-05: 99US models 2026-09-06: 99US models 2026-09-07: 99US models 2026-09-08: 99US models 2026-09-09: 99US models 2026-09-10: 99US models 2026-09-11: 99US models 2026-09-12: 99US models 2026-09-13: 99US models 2026-09-14: 99US models 2026-09-15: 99US models 2026-09-16: 99US models 2026-09-17: 99US models 2026-09-18: 99US models 2026-09-19: 99US models 2026-09-20: 99US models 2026-09-21: 99US models 2026-09-22: 99US models 2026-09-23: 99US models 2026-09-24: 99US models 2026-09-25: 99US models 2026-09-26: 99US models 2026-09-27: 99US models 2026-09-28: 99US models 2026-09-29: 99US models 2026-09-30: 99US models 2026-10-01: 99US models 2026-10-02: 99US models 2026-10-03: 99US models 2026-10-04: 99US models 2026-10-05: 99US models 2026-10-06: 99105All models 2026-07-13: 100All models 2026-07-14: 100All models 2026-07-15: 100All models 2026-07-16: 100All models 2026-07-17: 100All models 2026-07-18: 100All models 2026-07-19: 100All models 2026-07-20: 100All models 2026-07-21: 100All models 2026-07-22: 100All models 2026-07-23: 100All models 2026-07-24: 100All models 2026-07-25: 100All models 2026-07-26: 100All models 2026-07-27: 100All models 2026-07-28: 100All models 2026-07-29: 100All models 2026-07-30: 100All models 2026-07-31: 100All models 2026-08-01: 100All models 2026-08-02: 100All models 2026-08-03: 100All models 2026-08-04: 100All models 2026-08-05: 100All models 2026-08-06: 100All models 2026-08-07: 100All models 2026-08-08: 100All models 2026-08-09: 100All models 2026-08-10: 100All models 2026-08-11: 100All models 2026-08-12: 100All models 2026-08-13: 100All models 2026-08-14: 100All models 2026-08-15: 100All models 2026-08-16: 100All models 2026-08-17: 100All models 2026-08-18: 100All models 2026-08-19: 100All models 2026-08-20: 100All models 2026-08-21: 106All models 2026-08-22: 106All models 2026-08-23: 105All models 2026-08-24: 105All models 2026-08-25: 105All models 2026-08-26: 105All models 2026-08-27: 105All models 2026-08-28: 105All models 2026-08-29: 105All models 2026-08-30: 105All models 2026-08-31: 105All models 2026-09-01: 105All models 2026-09-02: 105All models 2026-09-03: 105All models 2026-09-04: 105All models 2026-09-05: 105All models 2026-09-06: 105All models 2026-09-07: 105All models 2026-09-08: 105All models 2026-09-09: 105All models 2026-09-10: 105All models 2026-09-11: 105All models 2026-09-12: 105All models 2026-09-13: 105All models 2026-09-14: 105All models 2026-09-15: 105All models 2026-09-16: 105All models 2026-09-17: 105All models 2026-09-18: 105All models 2026-09-19: 105All models 2026-09-20: 105All models 2026-09-21: 105All models 2026-09-22: 105All models 2026-09-23: 105All models 2026-09-24: 105All models 2026-09-25: 105All models 2026-09-26: 105All models 2026-09-27: 105All models 2026-09-28: 105All models 2026-09-29: 105All models 2026-09-30: 105All models 2026-10-01: 105All models 2026-10-02: 105All models 2026-10-03: 105All models 2026-10-04: 105All models 2026-10-05: 105All models 2026-10-06: 105
Chinese modelsUS modelsAll models

Base 2026-07-13 = 100 · updated 2026-10-06 · The median US model costs 2.6× the median Chinese model.

MethodologyChain-linked geometric mean · base 2026-07-13 = 100

Data is a daily snapshot of each model’s official API list price; the price used is the blended price = (3 × input + 1 × output) / 4, in USD per 1M tokens. The index is a chain-linked geometric mean: each day it takes only models listed on both that day and the previous one, computes the geometric mean of their price ratios and multiplies it onto the previous day’s index, so adding or retiring models never moves it; only price changes of listed models do. USD is the stable basis (a model quoted in CNY keeps a constant USD figure at the daily rate). A falling index means the same thing got cheaper; it says nothing about quality. For quality, see the cost-vs-benchmark section below.

COST VS QUALITY

Cost vs. quality: who is cheapest at a given scoreGreen dots are the frontier: no cheaper model scores higher; grey dots are the rest

1225385063$1.00$10Blended price / 1M tokens (log scale)AA Intelligence IndexGPT-5.6 Terra: $4.50 · 42.0GPT-5.5: $11 · 26.0Claude Haiku 4.5: $2.00 · 17.0Gemini 3.5 Flash-Lite: $0.85 · 22.0Gemini 3.1 Pro: $4.50 · 30.0DeepSeek V4 Pro: $1.98 · 36.0DeepSeek V4.1 Flash: $0.53 · 39.0Kimi K3: $5.96 · 44.0Kimi K2.7 Code: $1.71 · 26.0GLM-5.3: $2.15 · 45.0MiniMax M3: $0.53 · 29.0GPT-6 Astra: $20 · 53.0Claude Fable 5.1: $20 · 53.0Gemini 3.8 Flash: $1.50 · 41.0Qwen3.8-Max: $3.00 · 45.0Muse Spark 1.3: $2.00 · 48.0GPT-5.3 Codex: $4.81 · 33.0Grok 4.7: $3.00 · 46.0Claude Opus 5.5: $8.00 · 58.0GPT-6 Luna: $0.20 · 38.0Step 5 Preview: $1.53 · 44.0Qwen3.8 2.4T A95B: $3.00 · 40.0Qwen3.8 27B: $1.12 · 34.0MiMo-V2.6-Pro: $0.54 · 46.0Solar Pro 4: $0.16 · 28.0MiMo-V2.6-Flash: $0.17 · 38.0Claude Sonnet 5.5: $4.00 · 56.0GPT-6.1 Sol: $4.00 · 52.0GLM-5.3-Flash: $0.24 · 42.0Solar Pro 4MiMo-V2.6-FlashGLM-5.3-FlashMiMo-V2.6-ProMuse Spark 1.3Claude Sonnet 5.5Claude Opus 5.5

Cheapest model reaching a score level (AA Intelligence Index)

  • At least 95% of the top score: Claude Sonnet 5.5 · $4.00
  • At least 90% of the top score: Claude Sonnet 5.5 · $4.00
  • At least 80% of the top score: Muse Spark 1.3 · $2.00
  • At least 70% of the top score: GLM-5.3-Flash · $0.24
  • At least 60% of the top score: MiMo-V2.6-Flash · $0.17

Price-score frontier · AA Intelligence Index

ModelBlended / 1M tokensAA Intelligence Index% of top score
Solar Pro 4Upstage$0.1628.048%
MiMo-V2.6-FlashXiaomi$0.1738.066%
GLM-5.3-FlashZhipu AI$0.2442.072%
MiMo-V2.6-ProXiaomi$0.5446.079%
Muse Spark 1.3Meta$2.0048.083%
Claude Sonnet 5.5Anthropic$4.0056.097%
Claude Opus 5.5Anthropic$8.0058.0100%

Price-score frontier · LiveBench

ModelBlended / 1M tokensLiveBench% of top score
GPT-6 LunaOpenAI$0.2072.086%
GPT-5.6 LunaOpenAI$0.4573.688%
DeepSeek V4.1 FlashDeepSeek$0.5381.197%
Muse Spark 1.3Meta$2.0081.698%
GPT-6.1 SolOpenAI$4.0081.698%
Claude Opus 5.5Anthropic$8.0083.2100%
Claude Fable 5.1Anthropic$20.0083.4100%
GPU RENTAL

GPU cloud rentalPer hour, per GPU

GPUHourly rateSamplesAs of
B300 288GB288 GB$9.9412026-10-05
B200 180GB180 GB$6.25–$9.3852026-10-06
H200 141GB141 GB$4.21–$5.1242026-10-06
H100 SXM5 80GB80 GB$1.79–$4.0052026-10-06
L40S 48GB48 GB$0.60–$0.82132026-10-06
RTX 5090D 24GB24 GB$0.43–$0.54302026-10-06
RTX 4090D 24GB24 GB$0.31–$0.43302026-10-06

Prices come from the public vast.ai market API (international cloud GPUs); they do not represent China-market purchases or domestic cloud prices. Daily snapshots began 2026-10-06; a GPU rental index appears here once 3+ days are stored.

DATA

Download and cite

Free to cite: please credit “TANDIU, tandiu.cn” and link to this page. History files are append-only; past rows are never edited.