Skip to content
TokenLensCompare plans

Most AI coding plans won’t tell you what you’re buying.

We track 25 plans across 9 platforms and price every one we can. Most we can’t — the vendor publishes no limit that converts to tokens, so we say that rather than inventing a number.

Data verified 22 Sep 2026

Paid plans tracked

18

3
publish a limit you can convert to tokens — all of them GitHub Copilot
15
sell an allowance nobody will quantify

Pay-per-token API rates, for contrast, run $0.55 to $24.00 per 1M — every one of them published.

Everyone sells hours of coding. 3 of 25 plans will tell you how many tokens that is.

Platform comparison

Normalised cost across all plans. "Your cost/1M" spreads the monthly price over the estimated token cap; "Model rate/1M" is the underlying pay-per-token rate at a 65% input / 35% output split.

Every tracked plan, with its monthly price, estimated monthly token cap, the resulting cost per 1M tokens, the underlying model’s pay-per-token rate, and how confident we are in the figures.
PlanPrice/moToken cap/moYour cost/1MModel rate/1MModelsConfidence
CursorIDE4 plans
Cursor HobbyFreenot published—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Cursor Pro$20not published—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Cursor Pro Plus$603× a base plan
base cap not published
—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Cursor Ultra$20010× a base plan
base cap not published
—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Claude CodeHybrid3 plans
Claude Pro$20not published—$24.00
Claude Fable 5.1Claude Opus 5.5Claude Sonnet 5Claude Haiku 4.5
estimated confidence: inferred from marketing claims, not published numbers
Claude Max 5x$1005× a base plan
base cap not published
—$24.00
Claude Fable 5.1Claude Opus 5.5Claude Sonnet 5Claude Haiku 4.5
partial confidence: partly published, partly inferred
Claude Max 20x$20020× a base plan
base cap not published
—$24.00
Claude Fable 5.1Claude Opus 5.5Claude Sonnet 5Claude Haiku 4.5
partial confidence: partly published, partly inferred
GitHub CopilotIDE4 plans
Copilot FreeFreenot published—$4.80
Claude Sonnet 5GPT-5.6 SolGemini 3.1 Pro Preview
sourced confidence: taken directly from the provider's published figures
Copilot Pro$10
3.1M
published $ credit pool ÷ model rate
$3.20$4.80
Claude Sonnet 5GPT-5.6 SolGemini 3.1 Pro Preview
sourced confidence: taken directly from the provider's published figures
Copilot Pro+$39
14.6M
published $ credit pool ÷ model rate
$2.67$4.80
Claude Sonnet 5GPT-5.6 SolGemini 3.1 Pro Preview
sourced confidence: taken directly from the provider's published figures
Copilot Max$100
41.7M
published $ credit pool ÷ model rate
$2.40$4.80
Claude Sonnet 5GPT-5.6 SolGemini 3.1 Pro Preview
sourced confidence: taken directly from the provider's published figures
OpenAI CodexHybrid4 plans
ChatGPT Go$8not published—$4.80
GPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.3-Codex
estimated confidence: inferred from marketing claims, not published numbers
ChatGPT Plus$20not published—$24.00
GPT-6 AstraGPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.3-Codex
estimated confidence: inferred from marketing claims, not published numbers
ChatGPT Pro 5x$1005× a base plan
base cap not published
—$24.00
GPT-6 AstraGPT-6 Astra ProGPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.3-Codex
partial confidence: partly published, partly inferred
ChatGPT Pro 20x$20020× a base plan
base cap not published
—$24.00
GPT-6 AstraGPT-6 Astra ProGPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.3-Codex
partial confidence: partly published, partly inferred
Antigravity FreeFreenot published—$5.50
Gemini 3.1 Pro PreviewGemini 3.8 Flash
estimated confidence: inferred from marketing claims, not published numbers
Google AI Pro$20not published—$5.50
Gemini 3.1 Pro PreviewGemini 3.8 Flash
estimated confidence: inferred from marketing claims, not published numbers
Google AI Ultra 5x$1005× a base plan
base cap not published
—$5.50
Gemini 3.1 Pro PreviewGemini 3.8 Flash
estimated confidence: inferred from marketing claims, not published numbers
Google AI Ultra 20x$20020× a base plan
base cap not published
—$5.50
Gemini 3.1 Pro PreviewGemini 3.8 Flash
estimated confidence: inferred from marketing claims, not published numbers
Devin (Cognition)Hybrid3 plans
Devin FreeFreenot published—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Devin Pro$20not published—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Devin Max$20010× a base plan
base cap not published
—$4.80
Claude Sonnet 5Claude Opus 5GPT-5.6 SolGemini 3.1 Pro Preview
estimated confidence: inferred from marketing claims, not published numbers
Anthropic APIAPI1 plan
Anthropic APIpay-per-tokenunmetered—$24.00
Claude Fable 5.1Claude Opus 5.5Claude Sonnet 5Claude Haiku 4.5
sourced confidence: taken directly from the provider's published figures
OpenAI APIAPI1 plan
OpenAI APIpay-per-tokenunmetered—$24.00
GPT-6 AstraGPT-6 Astra ProGPT-5.6 SolGPT-5.6 TerraGPT-5.6 LunaGPT-5.3-Codex
sourced confidence: taken directly from the provider's published figures
Google AI APIAPI1 plan
Google AI APIpay-per-tokenunmetered—$5.50
Gemini 3.1 Pro PreviewGemini 3.8 Flash
sourced confidence: taken directly from the provider's published figures

Scenario calculator

Adjust your usage pattern to see how much you'd spend on API vs subscription.

Break-even chart

Where API cost crosses each subscription price as hours per day increases.

Dev-Hour Equivalents

Cost per hour of AI-assisted development work, by task type and model.

Cost of one Dev-Hour Equivalent for each benchmark task, by model.
BenchmarkToken budgetAnthropic: Claude Fable 5.1Anthropic: Claude Haiku 4.5Anthropic: Claude Opus 5Anthropic: Claude Opus 5.5Anthropic: Claude Sonnet 5Google: Gemini 3.1 Pro PreviewGoogle: Gemini 3.8 FlashOpenAI: GPT-5.3-CodexOpenAI: GPT-5.6 LunaOpenAI: GPT-5.6 SolOpenAI: GPT-5.6 TerraOpenAI: GPT-6 AstraOpenAI: GPT-6 Astra Pro
Dev-Hour Equivalent: Light Debug
Fix a small bug with minimal context; includes quick diagnosis + patch + unit test.
25K in
15K out
$1.0000$0.1000$0.5000$0.4000$0.2000$0.2300$0.0750$0.2537$0.0230$0.2000$0.2300$1.0000$1.0000
Dev-Hour Equivalent: Medium Feature
Implement a medium feature across multiple files with validation and tests.
80K in
60K out
$3.8000$0.3800$1.9000$1.5200$0.7600$0.8800$0.2850$0.9800$0.0880$0.7600$0.8800$3.8000$3.8000
Dev-Hour Equivalent: Large Refactor
Refactor multiple modules with structure changes and regression tests.
180K in
120K out
$7.8000$0.7800$3.9000$3.1200$1.5600$1.8000$0.5850$1.9950$0.1800$1.5600$1.8000$7.8000$7.8000
Dev-Hour Equivalent: Long-Context Review
Review a large codebase slice (long context) and propose changes with patch notes.
220K in
70K out
$5.7000$0.5700$2.8500$2.2800$1.1400$1.2800$0.4275$1.3650$0.1280$1.1400$1.2800$5.7000$5.7000

Most capability per dollar

Epoch Capabilities Index divided by blended API cost. Refreshed 30 Sep 2026.

ModelECICost / 1MPoints per $
Mistral: Mistral Nemo118.6$0.025191.7
OpenAI: gpt-oss-20b137.8$0.043190.0
Qwen: Qwen3.7 Flash144.6$0.072224.9
Mistral: Mistral Small 3127.1$0.062100.2
Meta: Llama 3.1 8B Instruct116.5$0.061925.3
Full leaderboard

Use TokenLens in your agent

Install the MCP server so Claude Code, Cursor, or Windsurf can answer pricing questions inline. Four steps, about two minutes.

Once installed, ask your agent in plain language. It picks the right tool and answers from live TokenLens data.

flikQ/tokenlens-mcp
Example conversation
YouWhich AI plan should I be on? I code about 5 hours a day.
Agentrecommend_planLet me check TokenLens for your usage profile…
AnswerCursor Pro ($20/mo) saves about $38/mo against the API at 5h/day on Claude Sonnet. Break-even is 2.4h/day, so you are well past it.

You will need Node.js 18+, Bun (or npm) and Git.

  1. Clone the server

    Standalone — no monorepo required.

    git clone https://github.com/flikQ/tokenlens-mcp.git
    cd tokenlens-mcp
  2. Build it

    Compiles TypeScript to dist/index.js.

    bun install
    bun run build
  3. Connect your editor

    Replace /path/to/tokenlens-mcp with your actual clone path.

    Claude Code — one command

    claude mcp add tokenlens -- \
      env TOKENLENS_DATA_URL=https://tokenlens.flikq.dev/api/data \
      node /path/to/tokenlens-mcp/dist/index.js

    Cursor, Windsurf or Claude Desktop — add to ~/.cursor/mcp.json or ~/.codeium/windsurf/mcp_config.json

    {
      "mcpServers": {
        "tokenlens": {
          "command": "node",
          "args": ["/path/to/tokenlens-mcp/dist/index.js"],
          "env": {
            "TOKENLENS_DATA_URL": "https://tokenlens.flikq.dev/api/data"
          }
        }
      }
    }
  4. Check it works

    Restart your editor, then ask your agent:

    Use the tokenlens recommend_plan tool. I code 3h/day on claude-sonnet.

Tools it adds

compare_plans
Ranks every plan by cost per 1M tokens for a model
run_scenario
API against subscription cost for your usage pattern
break_even
Hours per day where a plan starts paying for itself
recommend_plan
Single best-value plan within a budget

Methodology

Every figure here comes from a documented calculation — how token caps are estimated from weekly, daily and multiplier limits, how cost per 1M tokens is blended, and what each confidence tier means.

Read the methodology