Calculate your LLM
API costs.

Enter token usage to estimate model costs and inspect billing conditions. Calculate without signing in; save to continue on other devices.

Calculate API costs Find a model quote first

10 quote sources223 modelsLedger updated

How to calculate Token costs →

01 / API PRICING

Model API pricing ledger

Choose USD or CNY and estimate costs for 1–5 models using your input and output volumes.

Loading data
LIVE COST WORKSPACE

API prices and usage costs,
on the same scale.

Explore observed quotes, then estimate costs using your input and output volumes.

Same usage, different costs

USD · estimate
1M input + 250K output tokens · Selected model preview
01 / EXPLORE

Start with a model, explore your options

Browse quote sources and add models to compare. Units: USD per million tokens.

Model / billing conditionsInput priceOutput priceCache readCompare
02 / COMPARE

Compare differences on the same scale

Select 1–5 models and enter usage to estimate costs with current quotes.

USD / selected usage
Input costOutput costLinear scale

Links include models, billing conditions and usage assumptions. Opening a link recalculates with current quotes. Do not include sensitive business data.

Manage follows and reminders
Exchange rateCNYManual estimate · not a live exchange rate
Original quotes and sources are retained on model detail pages
MaaS Daily / Daily collection from eight official pricing sources.

Inspect original quotes by billing conditions. Source URLs are shown for verification.

Start with an observed model quote

Observed quotes; conditions and dates remain separate. Not a cheapest-model ranking.

ModelInput quoteOutput quoteContinue
02 / GPU MARKET

GPU rental market

Domestic platform reference prices, August 2026. Monthly costs use a square-root scale to show both consumer and data-center GPUs.

RTX 4090Entry-level inference, graphics and fine-tuning
Hourly¥1.24 ~ 2.8Monthly¥1000 ~ 2800
Suanjia Cloud / AutoDL and others
A100 80GMid-scale training and fine-tuning
Hourly¥≈ 3.6Monthly¥4000 ~ 15000
AutoDL / Jizhisuan
H100 80GHigh-end training
Hourly¥≈ 20Monthly¥50000 ~ 88000
InfoQ reviews / market reports
H200High-end training
HourlyNot availableMonthly¥72000 ~ 75000
Community market reports, July 2026
Ascend 910CDomestic training and inference alternative
HourlyNot availableMonthly¥≈ 60000
Community market reports, July 2026
Basis

Monthly estimates assume 24/7 operation. Prices vary with supply, demand and facility configuration; verify current platform quotes.

03 / CHIP SUPPLY

AI chip developments

Public information snapshot, September 1, 2026. Business and product milestones are grouped by vendor; figures with different definitions do not form a single ranking.

01

Huawei Ascend

2 signals
2025 Full-year shipments812,000 chips

Over 70% share in government, state-owned enterprise and telecom markets

Chuanguan News, August 2026
Share of domestic compute shipmentsNearly half

41% of domestic AI accelerator deliveries in 2025 (IDC)

Southern Finance, July 2026
02

Cambricon

3 signals
2025 Chip sales117,400 cards (+201.57%)

Production: 127,700 cards (+409.84%)

Securities Times, July 2026
2026 H1 revenueCNY 5.996 billion (+108.13%)

Net profit: CNY 2.311 billion (+122.61%); first STAR Market company to reach a CNY 1 trillion valuation

Dahe Financial Cube, September 2026
Siyuan 690Mass production in early 2026

Next-generation flagship cloud AI chip with dual-die packaging

Securities Times, July 2026
03

Enflame

1 signals
IPO progressListed in September 2026

Four leading domestic AI chip companies enter capital markets

Guancha, August 2026