What it is
Claude Code is Anthropic's agentic coding tool. You describe a task in plain language and it reads the codebase, plans, edits files across the project, runs tests and commands, and can commit and open pull requests. It started as a terminal program and now also runs as a VS Code and JetBrains extension, in the Claude desktop app, on the web at claude.ai/code, and from the Claude mobile apps and Slack.
It is built to be scripted and extended: a CLAUDE.md file holds project instructions, skills package repeatable workflows, hooks run shell commands around its actions, and MCP connects it to outside tools. It can spawn subagents, run background sessions, and run scheduled "routines" in the cloud. The Agent SDK exposes the same engine for building your own agents.
Since version 2.1.280 on September 22, 2026, the default model is Claude Opus 5.5, which has a 1M-token context window, and Opus is now the default on Pro and Team Standard plans as well as Max.
Our review
Claude Code is the coding agent its rivals are measured against, and the reason is the harness as much as the model. It works in your real repository with your real shell, and the same engine runs in the terminal, VS Code, JetBrains, the desktop app, the browser and Slack; a cloud task started on the web can be pulled into a local terminal with --teleport. Skills, hooks, plugins, subagents and the Agent SDK give a platform team more control over its behaviour than most rivals expose, and managed settings let an organisation enforce that control on every machine.
Since 22 September 2026 the default model is Opus 5.5. Anthropic prices it at $4 per million input tokens and $20 per million output, 20% below Opus 5, and claims about 40% lower cost on typical work because it also spends fewer tokens per task. The changelog moved Pro and Team Standard from Sonnet to Opus by default the same day. For a $20 subscriber that is a real upgrade, and also a quicker route to the usage cap.
The number a buyer should plan around sits in Anthropic's own cost guide: across enterprise deployments Claude Code averages about $13 per developer per active day and $150 to $250 per developer per month on API billing, with 90% of users under $30 a day. Subscriptions publish no message counts at all. Usage resets on a rolling five-hour window with weekly caps, and it is shared with Claude chat and Cowork, so a heavy coding afternoon eats the allowance a developer would otherwise spend on writing. Agent teams, off by default, use roughly seven times the tokens of a normal session when teammates run in plan mode.
The security defaults deserve a careful read. Auto mode, in which a separate classifier model approves actions instead of the developer, is now the starting permission mode for interactive terminal and VS Code sessions; Manual mode, which starts read-only and asks before edits or commands, has to be chosen or enforced. Session transcripts sit in plaintext under ~/.claude/projects for 30 days. On Pro and Max, code is used for training if the account's training setting is on, with five-year retention; Team, Enterprise and API traffic is not trained on and is kept for 30 days, and zero data retention is an Enterprise option Anthropic enables per organisation.
Where it falls behind is choice. Cursor, Cline and GitHub Copilot let a team pick GPT or Gemini from a menu; Claude Code reaches Amazon Bedrock, Google Cloud and Microsoft Foundry, but only to run Claude. For a team already standardised on Anthropic that lock-in costs nothing, and for one hedging across model vendors it is the whole decision.
Where it earns its place
A platform team of eight engineers maintaining a Python monorepo
Automate dependency bumps and flaky-test fixes in CI
Run Claude Code headless with -p in the pipeline, constrained by hooks and deny rules, and bill it through a Console workspace with its own spend limit and rate limit, so a runaway loop cannot drain the production API quota.
An engineering manager rolling out an agent to 40 developers
Give every developer Claude Code without an open-ended bill
Team Premium seats at $100 a month on annual billing come to $4,000 a month with a fixed ceiling; usage past the seat allowance bills at API rates only if usage credits are switched on, and spend limits can be set per member. Standard seats at $20 suit lighter users but run out sooner.
A freelance developer on the $20 Pro plan
Ship client features from the terminal without a second subscription
Pro covers occasional sessions, but Opus as default and the shared five-hour window mean daily agent use often ends in a move to Max at $100, which also includes $100 a month in API credits, according to users on Reddit and Hacker News.
A security lead at a 200-person fintech
Approve an AI agent that edits production code
Commercial plans mean no training and 30-day retention; zero data retention needs Enterprise and Anthropic's sign-off. Enforcing Manual mode, the bash sandbox and an approved MCP list through managed settings is the realistic baseline, since Auto mode is the default.
Before you commit
- Setup
- Install the CLI or the VS Code or JetBrains extension and sign in with a paid Claude account or a Console API key; a first task runs within minutes, and a CLAUDE.md file holds project rules.
- Integrations
- MCP servers, GitHub for pull requests and cloud sessions, Slack, and Claude via Amazon Bedrock, Google Cloud's Agent Platform or Microsoft Foundry; OpenTelemetry export gives per-user token and cost metrics.
- Data and privacy
- Team, Enterprise and API: no training, 30-day retention, zero retention on approved Enterprise accounts. Pro and Max: training if the setting is on (five-year retention), 30 days if off. Local transcripts are plaintext for 30 days. SOC 2 Type 2 and ISO 27001 per Anthropic's trust centre.
- Cost at scale
- Anthropic reports an enterprise average of about $13 per developer per active day, or $150 to $250 a month, on API billing; 50 developers on Team Premium seats cost $5,000 a month on annual billing before any usage credits.
- Limits
- Subscriptions reset on a rolling five-hour window with weekly caps and no fixed message count; for API teams of 100 to 500 users Anthropic suggests 15k to 20k tokens per minute per user.
What's good
- Opus 5.5 by default, at $4/$20 per million tokens on the API
- Same agent in terminal, IDEs, desktop, web and Slack, with --teleport between them
- Hooks, skills, plugins, subagents and managed settings for organisation-wide control
- Cloud sessions run in isolated VMs; GitHub credentials never enter the VM
- Included from the $20 Pro plan
What's not
- Runs Claude models only, even on Bedrock, Google Cloud or Foundry
- Plan usage is shared with Claude chat, with no published message counts
- Auto mode, where a classifier approves actions, is the default in the terminal
- Pro and Max code is used for training when the setting is on, kept five years
- Not on the free Claude plan
Best for
Developers who want a scriptable, highly customisable coding agent and are happy standardising on Claude models.
Alternatives
Head to head
Claude Code: common questions
- Is Claude Code worth it?
- Claude Code sets the standard for terminal coding agents in 2026; the price of that standard is a single model vendor and a usage meter you cannot see. We rate it 4.7 out of 5.
- How much does Claude Code cost?
- Included in every paid Claude plan: Pro $20/month ($17/month billed yearly); Max 5x $100/month or Max 20x $200/month; Team Standard $25/seat/month ($20 yearly) and Premium $125/seat/month ($100 yearly); Enterprise $20/seat/month plus usage at API rates. Not on the free plan. Can also run on API pay-as-you-go.
- What are the downsides of Claude Code?
- Runs Claude models only, even on Bedrock, Google Cloud or Foundry; plan usage is shared with Claude chat, with no published message counts; auto mode, where a classifier approves actions, is the default in the terminal; pro and Max code is used for training when the setting is on, kept five years; not on the free Claude plan.
- Who is Claude Code best for?
- Developers who want a scriptable, highly customisable coding agent and are happy standardising on Claude models.
- How much does Claude Code cost for a team or in production?
- Anthropic reports an enterprise average of about $13 per developer per active day, or $150 to $250 a month, on API billing; 50 developers on Team Premium seats cost $5,000 a month on annual billing before any usage credits.
- Is Claude Code safe for company data?
- Team, Enterprise and API: no training, 30-day retention, zero retention on approved Enterprise accounts. Pro and Max: training if the setting is on (five-year retention), 30 days if off. Local transcripts are plaintext for 30 days. SOC 2 Type 2 and ISO 27001 per Anthropic's trust centre.
- What are the best Claude Code alternatives?
- The closest alternatives we have reviewed are OpenAI Codex, Cursor, Google Antigravity (with Gemini CLI and Jules) and GitHub Copilot.
Answers checked Oct 23, 2026.
Make Claude Code? Show this rating on your site
