Rankings by use case

Best LLM APIs for math and reasoning (August 2026)

The models rated highest on math and logical reasoning, restricted to what FastMetal serves today.

Ranked by ELO on the public arena's Math board, restricted to models FastMetal serves. The order updates with the leaderboard.

Links
1Claude Fable 5anthropic logoAnthropic#1 · 1,539¥1,787¥8,9351,000,000DetailsCompare
2Claude Opus 5anthropic logoAnthropic#2 · 1,533¥893.5¥4,467.51,000,000DetailsCompare
3Gemini 3.5 Flashgoogle logoGoogle#3 · 1,520¥268.05¥1,608.31,048,576DetailsCompare
4Claude Opus 4.6anthropic logoAnthropic#7 · 1,503¥850¥4,2001,000,000Details
5Grok 4.5x-ai logoxAI#8 · 1,497¥357.4¥1,072.2500,000DetailsCompare
6Claude Opus 4.7anthropic logoAnthropic#13 · 1,492¥840¥4,2001,000,000Details
7GPT-5.6 Terraopenai logoOpenAI#14 · 1,490¥357.4¥2,144.41,050,000DetailsCompare
8Qwen3.7 Maxqwen logoQwen#15 · 1,490¥263.58¥790.751,000,000DetailsCompare
9GLM 5.1z-ai logoZ.ai#18 · 1,482¥250.18¥786.28204,800Details
10Kimi K2.6moonshotai logoMoonshot#19 · 1,480¥169.77¥714.8262,144DetailsCompare

Click a column heading to re-sort; the table opens in this category's ranking order. Prices are yen per 1M tokens before tax; context is in tokens.

Frequently asked questions

What is this ranking based on?
Ranked by ELO on the public arena's Math board, restricted to models FastMetal serves. The order updates with the leaderboard.
How is pricing decided?
Each model has a yen price per 1M tokens (before tax), drawn from a prepaid balance as you use it. There is no subscription or monthly fee.
Can I use the top models with one API key?
Yes. Every model on this page is served from FastMetal's OpenAI-compatible endpoint under one API key; switching is a change to the model string.

Every model above runs on one API key

Create an account and add credit to call these models from the browser chat and the API. No monthly fee.

Other rankings