How to read the numbers (re-checked 2026-10-09): the revenue figures now come from Anthropic’s own announcement (Series G, published 2026-02-12); they are vendor-reported, not audited, and this site did not verify them independently. The earlier secondary-only figures (OpenAI Codex ~$1B; the 54%/46% survey shares) could not be anchored to a first-party source and were removed.
The headline: Anthropic’s own announcement (2026-02-12) says Claude Code’s run-rate revenue passed $2.5 billion — more than doubling since the beginning of 2026 — with weekly active users also doubling since January 1. Auto Mode, introduced on 24 March 2026 as a research preview, became generally available in Claude Code for all users on 10 July 2026 (Anthropic’s own page); it moves permission decisions into a classifier and is documented for Amazon Bedrock, Google Cloud’s Agent Platform and Microsoft Foundry.
The market numbers
The reported figures, in order of magnitude:
- Claude Code run-rate revenue: over $2.5 billion (Anthropic, Series G announcement, 2026-02-12), more than doubled since the beginning of 2026; weekly active users doubled since 1 January 2026, and Anthropic says enterprise customers contribute more than half of Claude Code revenue.
- ~4% of all public GitHub commits were estimated to be authored by Claude Code — double the share of one month earlier. Anthropic cites the analysis; it is an estimate, not a metric.
- What was removed: the earlier “Codex ~$1B” and the 54% / 46% survey shares came only from secondary coverage with no first-party anchor, so this page no longer repeats them.
These are Anthropic’s own disclosure figures, not audited results. The direction is still unambiguous: a terminal tool nobody had heard of in 2024 is now the reference point for the category.
Why Claude Code won
Our reading of the available reporting is that the driver is less raw model quality than workflow data.
The second factor is the workflow gate, not the model. Claude Code turned plan → review → approve → execute into the product itself: permission rules, hooks, sandboxes, and review gates are part of the tool, not add-ons. That is why enterprises pay for it rather than routing around a raw API.
Auto Mode GA: the boundary moves
Auto Mode: what actually changed. Anthropic introduced Auto Mode on 24 March 2026 as a research preview and updated the same page on 10 July 2026: Auto Mode is generally available in Claude Code for all users, and the current permission-mode documentation lists it for Amazon Bedrock, Google Cloud’s Agent Platform and Microsoft Foundry. Before each tool call runs, a classifier reviews it for potentially destructive actions — mass file deletion, sensitive-data exfiltration, malicious code execution — and actions it blocks can still be sent back to the user for authorisation. Anthropic documents it as a safer long-running alternative to --dangerously-skip-permissions, not as a removal of every gate.
Our analysis: permission prompts are the biggest friction in agentic coding, and moving them into a classifier changes where review happens. A control a human used to see now runs before execution — better than nothing, but the boundary moves into your environment: approvals, sandboxing and network egress controls enforced outside the agent’s loop.
The governance question is the part worth scrutinising: Auto Mode replaces a prompt a human sees with a classifier that runs before actions execute, which changes where the review boundary sits.
What the documentation does not say is that the boundary disappears: blocked actions can still be escalated to the user, and the classifier only covers the actions its rules review. The practical question for teams is unchanged — who reviews the changes the classifier lets through.
The cost side of the win
The cost side, as reported. Secondary market coverage (Tiger Brokers) describes customer cost concerns after the move to usage-based billing; no first-party cost data was re-read for this update, so treat those cost claims as reported and unverified.
What teams should do differently
The rise of Claude Code changes the competitive picture but not the evaluation checklist. The same bounded process applies:
- Your code, your acceptance criteria. Market share is not a benchmark on your codebase. Run a bounded pilot against your own tasks, your own repos, your own review process.
- Keep a review gate outside the classifier. Auto Mode reduces routine prompts; it does not remove review. Risky actions are blocked and redirected, repeatedly blocked attempts eventually trigger a permission prompt, and the classifier itself can err in both directions. Keep approvals, sandboxing and network egress controls enforced outside the agent’s loop — the boundary is the product, for every agent, not just this one.
- Model per-task cost, not token price. With usage-based billing, the bill is a function of task design. Measure real tasks end to end before committing.
- Know the data path. Terminal agents with credentials run where your developers run. If that is unacceptable for your compliance context, that is a platform decision, not a tool decision — which is where self-hosted, managed environments and managed agent work change the answer.
The take
Claude Code’s dominance is real, and it is built on workflow data plus a genuine product insight: the gate is the product. Auto Mode at GA is that logic taken further: the review gate moves from a human prompt into a classifier that runs before execution. For engineering teams, the conclusion is the same one that comes out of every agent story this year: the model is not the boundary, the environment is. Whoever wins the market, the team that controls where agents run, what they can touch, and how their actions are reviewed will get more from any of them.