# aicodingpricing leaderboard optimization brief

Source: 孟健 Telegram DM, 2026-05-28

## Decision

把 AI Coding Model Leaderboard 做成 aicodingpricing.com 的核心内页，不单独起站。

Product positioning upgrade:
- From: AI model pricing table
- To: AI coding model decision engine

## Core positioning

Page concept:
AI Coding Model Leaderboard

Subhead:
Compare the best LLMs for coding agents by benchmark performance, API pricing, context window, speed, and real task cost.

Do NOT build a generic LLM leaderboard. Competing references already include LMArena, Artificial Analysis, Hugging Face, OpenRouter. This site should focus on AI coding model selection.

## Proposed URL matrix

Primary:
- /llm-leaderboard
- /best-llm-for-coding
- /coding-agent-cost-calculator

Expansion / supporting pages:
- /ai-coding-model-leaderboard
- /coding-model-benchmark
- /llm-api-pricing-comparison
- /claude-vs-gpt-for-coding
- /kimi-vs-qwen-vs-deepseek-coding
- /models/{model}

## Differentiation

This is not a comprehensive model intelligence ranking. It is a decision page for AI coding workflows.

Core dimensions:
- coding benchmarks: SWE-Bench, Terminal-Bench, LiveCodeBench, Aider polyglot, etc.
- agent capability: tool use, long-horizon execution, repo-level editing, browser/computer use
- price: input/output/cache/batch pricing
- context: max context and effective long-context reliability
- speed: TTFT, tokens/sec; use public data first
- total task cost: average trajectory tokens × retry count × price
- use-case filter:
  - coding agent
  - frontend generation
  - repo refactor
  - bug fixing
  - code review
  - test generation
  - Chinese coding workflow
  - low-cost agent

## P0 scope

Build a semi-automated data page, not a complex realtime ranking.

P0 table fields:
- model
- provider
- coding score
- input price
- output price
- cache price
- context
- best for
- caveat

P0 filters:
- cheapest
- best coding
- best long context
- best Chinese coding
- best open model
- best agent model

P0 recommender options:
- I’m building a coding agent
- I need cheapest good-enough model
- I need frontend generation
- I need repo-level refactor

Methodology requirements:
- Do not force incomparable benchmarks into one fake absolute score
- Mark unavailable public data as not disclosed
- Clearly separate token price and task cost
- Cite/attribute public benchmark and speed sources

## SEO/product rationale

Potential value:
- `llm leaderboard` has meaningful demand, but generic leaderboard is hard to win
- aicodingpricing already owns pricing intent
- leaderboard upgrades the site from price table to model procurement decision tool
- target query clusters:
  - best llm for coding
  - coding model leaderboard
  - claude vs gpt coding
  - llm api pricing
  - cheapest coding model
  - best model for coding agents

Decision chain:
1. User searches model price
2. User asks whether cheap model can code well
3. User enters leaderboard: performance × price × scenario
4. User enters calculator: monthly task cost
5. Conversion: affiliate/newsletter/API provider/self-owned tool

## Explicit NOT-DO

- No claim of being the most authoritative global leaderboard
- No realtime scraping of every provider in P0
- No self-developed benchmark in P0
- No generic intelligence ranking
- No forced absolute total ranking across GPT / Claude / Gemini / Kimi / Qwen / DeepSeek when evidence is not comparable
- No filling missing data without public support

## Minimum shippable version

Priority:
1. /llm-leaderboard
2. /best-llm-for-coding
3. /coding-agent-cost-calculator

Initial implementation should integrate with existing aicodingpricing data model and calculator where possible.