金石API vs Modelflare

One model on two relays: integrity, speed, price, cache and availability, plus intelligence adjusted for degradation and value for money, with advice for the priority you choose.

Model
Gemini
gemini-premium

Which to choose

Five dimensions

Further out is better. Integrity is the share of intelligence kept; the other four are places among all relays of this model.
IntegrityNot checkedSpeedNot checkedPriceCacheAvailabilityNot checked
金石APIModelflare

What matters most to you

For Balanced, choose 金石API

Main reasons

Integrity25%

Integrity: no data for at least one relay

Availability20%

Availability: no data for at least one relay

Value20%

金石API gives 23.3 points per unit of money, 730% more than Modelflare at 2.80

Speed15%

Speed: no data for at least one relay

Price15%

金石API costs ¥2.64 per session, 88% less than Modelflare at ¥21.9

Cache5%

Cache is close: 金石API 95%, Modelflare 95%

金石API

Stronger
Price¥2.64Value23.3
Weaker
No clear gap

Modelflare

Stronger
No clear gap
Weaker
Price¥21.9Value2.80

Key figures

Adjusted intelligence

Higher is better

Cost per coding session

2M input + 40k output tokens · Lower is better

Value for money

Intelligence points per ¥1 · Higher is better

All figures

Intelligence

Adjusted intelligence61.461.4
Intelligence keptNot checkedNot checked
Question accuracyNot checkedNot checked

Cost

Value for money23.32.80
Cost per coding session¥2.64¥21.9
Input price¥1.2¥9.96
Output price¥6¥49.8
Cache discount95%95%

Performance

Time to first token——
Availability, last 24 hours——
Samples——

Route

GroupGeminigemini-premium
Data securityLevel 0Level 4
Adjusted intelligence = the model's published intelligence index × intelligence kept. Intelligence kept = the relay's question accuracy ÷ the highest accuracy among checked relays of the same model; under 10 questions, no adjustment. Value = adjusted intelligence ÷ cost per coding session. Gaps under a threshold count as close. Advice weighs the metrics by the chosen priority and is a guide only.