Loading...
按推理能力排名的顶级 AI 模型
o3 is the strongest reasoning model available, scoring 80 on reasoning and 85 on math while carrying the full capability set, making it the choice when output quality matters more than cost.
Rankings based on public benchmark data. Prices in USD per 1M tokens (direct provider). Updated June 2026.
o3 sets the ceiling. Reasoning at 80, math at 85 — the highest marks in this lineup, and the gap to everything else is clear rather than marginal. When you need the best possible answer on a genuinely hard problem, this is the model.
It also brings the full capability set. Vision, function-calling, tool use, all present, with a 200K context window. That matters for reasoning because the hardest problems often involve images, structured data, or multi-step tool interaction. A model that reasons well but cannot call tools or read a diagram is constrained in ways o3 is not.
The cost is real. At $2/$8, o3 is not cheap, and reasoning models spend heavily on thinking tokens billed at output rates, so a hard problem can run up a sizable trajectory. Budget for that if you commit.
The alternative worth naming is DeepSeek R1, at 72 reasoning and 78 math for roughly a tenth of the price. For text-only reasoning at volume, R1 closes much of the gap and saves enormous sums. o3 earns its premium specifically when you need the top of the curve, vision, or tool calls — not for routine analytical work where R1 suffices.
o3 remains the ceiling for reasoning that touches tools, and nothing here closes that gap. R1 is the value play for pure-text chains—roughly the same thinking, none of the tool-calling tax.
Last updated June 2026