Model Comparison | 模型对比

Compare AI models on the dimensions that actually decide a choice — price, context window, capabilities — for a specific task, and make a clear recommendation.

围绕真正决定取舍的维度（价格、上下文、能力）对比多个 AI 模型，针对具体任务给出明确建议。

When to use | 适用场景

• The user asks: 「对比一下 … 和 …」「哪个模型更适合做 …」「which model should I use for …」
• Choosing a model for a project, a budget, or a specific workload
• Re-evaluating an existing choice against the current landscape

Workflow | 工作流程

1. Scope the task — what will the model do? (chat, coding, long docs, vision, agentic calls, …) What matters most: cost, latency, quality, context size, or ecosystem?
2. Pick candidate models — include the obvious candidates plus 1–2 strong alternatives. Only include models the user can realistically access today.
3. Gather fresh facts — never rely on memory:
   • Web search for current pricing & context windows (prices change often).
   • For Anthropic models, consult the official Claude API pricing/docs.
   • Note the date of your data source; flag anything that may have changed.
4. Build a comparison table with columns: Model | Provider | Price (in/out per M tokens) | Context | Strengths | Weaknesses.
5. Recommend one — with a one-line reason, plus the runner-up and when the runner-up wins.
6. If prices/limits are not available or contradictory, say so and give a qualitative comparison.

Output format | 输出格式

# ⚖️ Model comparison: <task> | 模型对比

| Model | Provider | $/M in | $/M out | Context | 适合 Strength |
|---|---|---|---|---|---|
| ... | ... | ... | ... | ... | ... |

**建议 Recommendation**: **<Model>** — 因为 / because <one-line>.
**备选 Runner-up**: <Model> — 当 / when you need <…>.
*数据截至 / data as of: 2026-08-29. Prices change; verify before purchase.*

Tips | 提示

• Always search for current numbers; pricing and context windows change frequently.
• Compare apples to apples: same unit (per million tokens), same provider pricing page.
• If the task is agentic / tool-use heavy, weigh tool-use reliability over raw benchmark scores.
• If the user has a budget, put cost first in the table and rank by it.
• Recommend, don't dump — a table with no verdict is only half the job.