Is Claude Code open source?

No. The github.com/anthropics/claude-code repository states "© Anthropic PBC. All rights reserved. Use is subject to Anthropic's Commercial Terms of Service" and does not contain the application source (checked 2026-08-14). context-code is Context's distribution of Hermes Agent, which is MIT-licensed and fully readable.

Does context-code require a subscription?

No. Claude Code requires a Pro ($20/mo), Max ($100 to $200/mo), Team, Enterprise, or Console account; Anthropic's free plan does not include it. context-code has no subscription: you pair it with a Context workspace and sessions bill as usage like any other workspace task.

Which is cheaper to run?

It depends on the model you choose, so do the arithmetic at your own volume. At vendor list rates read on 2026-08-14, a month of 40M input and 4M output tokens computes to $300 on Claude Opus 5 ($5/$25 per M), $120 on Claude Sonnet 5 ($2/$10), $180 on Kimi K3 ($3/$15), and $20.88 on DeepSeek V4 Pro ($0.435/$0.87). Claude Code can only spend at Claude rates or within its subscription limits; context-code can route the same month to any of these.

Can context-code run Claude models?

Yes. Claude sits in the same catalog as GPT, Gemini, and the open-weight models, and you pick the model when you start a session. Choosing context-code is not a vote against Claude; it is a vote against being limited to it.

Does context-code work on Windows?

Through WSL 2: install WSL, open the Linux shell, and run the same one-liner macOS and Linux use. Claude Code additionally offers native Windows installers today; if native Windows is a hard requirement, that is a genuine point in its favor.

Are open-weight models actually good enough for real work?

On Artificial Analysis' Intelligence Index v4.1.1 (read 2026-08-14), Kimi K3 scores 60, above Claude Opus 4.8 at 57 and a few points below Anthropic's newest models at 61 to 63, with a 1M-token context window, at roughly half the list rate of Anthropic's flagship tier. For many agent workloads that trade is worth making per task rather than never.

Comparisons

Context Code vs Claude Code

Anthropic's own CLI for Claude models, or a workspace-paired CLI that runs the whole catalog. Which fits comes down to how much of your work is Claude-only.

Talk to our team

Vendor sources last verified August 2026.

Sessions land in the workspace task list

Runs

Durable runbook runs with rubric scores and an append-only audit trail

Runbook	Model	Status	Eval	Last action
Weekly pipeline review	Claude	Completed	96 / pass	Deck exported (.pptx)
Fix the flaky auth test	Kimi K3 · context-code	Running	Scoring	context-code session
Vendor diligence memo	GPT	In review	91 / pass	14 sources cited
Support ticket triage	Gemini	Running	Scoring	Authorized: read:tickets

Claude Code is Anthropic's own CLI and the polish shows: native installers for macOS, Linux, and Windows with auto-updates, deep first-party integration, and Claude models at their best. It requires a paid Anthropic account: Pro at $20 a month, Max at $100 to $200, Team, Enterprise, or an API key. And it runs Claude models only. Its repository ships under a proprietary license.

context-code is Context's distribution of the MIT open-source Hermes Agent. It installs with one command from a checksum-verified channel, pairs with your workspace once, and then every session runs on your machine while appearing in the workspace task list: visible to the team, billed as usage on one account, no subscription. The model catalog spans Claude, GPT, Gemini, and open weights: on Artificial Analysis' Intelligence Index (v4.1.1, read 2026-08-14) Kimi K3 scores 60, above Claude Opus 4.8's 57 and a few points under Anthropic's newest, at open-weight list rates.

If your team wants Anthropic's first-party tool and works Claude-only, buy Claude Code. Choose context-code when model choice, open-weight economics, an auditable open base, or team-visible sessions matter more than vendor depth. Plenty of engineers run both.

Sources last verified August 2026.

Compared August 2026

Models

What can run the work

Model choice

Open-weight models

Pick the model per session

Frontier-class open weights

Cost

What it takes to start and what a month costs

Price of entry

Works without a subscription

Cheapest strong model, list rates

A month at 40M in + 4M out

Distribution and source

What you install and whether you can read it

License

Read the source

Install

Platforms

Team and workspace

Where sessions live and who can see them

Where sessions appear

One bill for the team

Device credentials

Requires

context-code

Usage-based task billing

Talk to our team

Claude, GPT, Gemini, Kimi, DeepSeek, GLM

Kimi K3, DeepSeek V4, GLM

✓

Kimi K3: 60 on AA's index (Opus 4.8: 57)

No subscription; usage-based

✓

DeepSeek V4 Pro $0.435/$0.87 per M

$20.88 on DeepSeek V4 Pro rates

MIT base (Hermes Agent)

✓

One-liner, SHA-256-verified channel

macOS, Linux, Windows via WSL 2

Your terminal + the workspace task list

Workspace task usage

Pairing code; revocable per-device keys

A Context workspace

Claude Code

Pro/Max $20 to $200 a month, or API

Visit Anthropic

Claude models

Between Claude models

Not applicable

Pro $20/mo, Max $100 to $200/mo, or API

Sonnet 5 $2/$10 per M (API)

$120 to $300 at Claude rates

Proprietary, commercial terms

Native installers, brew, WinGet, npm

macOS, Linux, native Windows

Your terminal

Per-seat plans or Console billing

Account login or API key

Pro, Max, Team, Enterprise, or Console

Where each one fits

Running both

Common in practice: engineers keep Claude Code where a Max plan already covers Claude-only work, and use context-code when a task fits a different model, needs open-weight economics, or should be visible in the workspace task list. Claude models are first-class citizens of the context-code catalog, so switching tools does not mean giving them up.

Choose Claude Code when

You want Anthropic's own tool.: First-party polish: native installers with auto-updates on macOS, Linux, and Windows, and new Claude capabilities land in Claude Code first. If the vendor relationship is the point, buy from the vendor.
Your work is Claude-only and already paid for.: If the team runs on Max or Enterprise plans anyway, Claude Code's marginal cost is zero and its Claude integration is the deepest available.
You need native Windows today.: Claude Code installs natively on Windows 10 1809+ via PowerShell, CMD, or WinGet. context-code currently runs on Windows through WSL 2.

Choose Context when

You want the whole catalog, not one vendor.: Pick Claude for one session and Kimi K3 or DeepSeek for the next, from the same catalog your workspace runs. Open weights now score frontier-class: Kimi K3 rates 60 on Artificial Analysis' index, above Claude Opus 4.8.
The cost math matters at your volume.: No subscription in front of the first token, and open-weight list rates run far below frontier ones: the same 40M-in/4M-out month is $20.88 at DeepSeek V4 Pro rates against $120 to $300 at Claude rates. Sessions bill as workspace task usage.
Sessions should be visible to the team.: Every context-code session lands in the shared task list under a revocable per-device key and bills to the workspace account, not to a personal subscription nobody can see.

Questions worth asking both vendors

Whichever way this comparison lands for your team, these are the questions that separate a good demo from a platform that holds up in production.

Which models can it run, today?: Not which models exist: which ones this CLI can actually start a session with. A Claude-only tool is the right answer only if your work is Claude-only.
Price a realistic month of your own tokens.: Take your team's real volume and compute it at each option's published rates, subscriptions included. Ask both vendors to show the arithmetic rather than an aggregate multiplier.
Can you read the source of the thing running in your shell?: An agent CLI executes commands on developer machines. An MIT base can be audited line by line; a proprietary binary is a trust decision about the vendor.
Where do sessions appear, and who can see them?: Terminal-only sessions are invisible to the team. Ask whether a lead can see what agent sessions ran this week without collecting screenshots.
How do you revoke one laptop's access?: Listen for per-device credentials that can be revoked individually, versus a shared account login where revocation means rotating everything.
How do installs and updates prove integrity?: Both tools should answer well: signed installers or checksum-verified channels. Make the mechanism explicit rather than assumed.

FAQs

Claude Code and Context, answered plainly

Something missing or out of date? Report it.

Sources

Competitor facts come from these vendor pages, last verified August 2026.

Choosing between them

Working together

Something missing or out of date? Report it.

Sources

Competitor facts come from these vendor pages, last verified August 2026.

See Context on your workflows

Bring one real use case and watch agents build it on the deployment model you need: managed, in your VPC, on-premises, or air-gapped.

Talk to our team

More comparisons: ChatGPT Enterprise · Claude · Microsoft 365 Copilot · Glean · Zapier Agents