Thinking in Code

Analyze a 7 MB file. Pay for 407 bytes.

Your agent writes a short script. Context Mode runs it, and only what it prints reaches the model.

Try it

Question: how many versions of react are on npm?

The file7,011,585 B
The model sees100,000 chars

Sent to the model

  • The first 1.4% of the file

Never read

  • The other 98.6%
Long-Context Data Analysis · Claude Haiku 4.5$0.175 → $0.06264% lower12 of 12 runs

Question: how many versions of react are on npm?

The script reads7,011,585 B
The model reads407 B

Sent to the model

  • The answer: 2,959 versions, the latest stable one, and their dates

Never sent

  • The 7 MB file itself
Long-Context Data Analysis · Claude Haiku 4.5$0.175 → $0.06264% lower12 of 12 runs
How it works

The script reads. The model reads the answer.

  1. 01The agent writes a script

    For example, count versions in a registry document.

  2. 02Context Mode runs it

    Public data in a sandbox, local files on your machine.

  3. 03Only the result comes back

    The script read 7,011,585 bytes. The model read 407.

Proof

Data analysis, run twice. Cost per task.

Claude Haiku 4.5

Direct100%
Context Mode36%

64% lower

Claude Opus 5.5

Direct$0.223
Context Mode$0.201

10% lower

Paired runs, 2026-09-29. Correct answers were equal in both arms. Method and sources →

Features

Two places to run.

Public data

Web pages and APIs, read in a sandbox that reaches only the sites you allow.

Your files

Files on your machine, read by your agent with its own permissions.

Checked by Protect

Every host a script calls meets your rules.

Use cases: Analyze large files

Keep raw data out of the prompt.

npx @context-mode/cli connects your agents. 1,000 requests free, no card.

Get started free

Your coding agent costs less and remembers your project.

Start free: 1,000 requests, every feature, no card. After you sign in, the console gives you one command to connect Claude Code or Codex.

Read the docs