AI dev tools 2026
Claude Code vs Cursor vs Codex vs Antigravity in October 2026: pricing and governance
Four tools matter to production teams this month: Claude Code from Anthropic, Cursor, Codex from OpenAI and Google's Antigravity CLI. Model quality changes every few weeks, so this page compares what stays checkable: the plans, the prices and the admin controls each vendor documents. Every figure below was read from the vendor's own page on 19 October 2026.
Plans and prices side by side
| Tool | Entry | Heavy individual use | Team and enterprise |
|---|---|---|---|
| Claude Code | Not on the Free plan. Pro: $20 a month, or $17 on annual billing | Max: from $100 a month, with 5x or 20x the usage of Pro | Team: standard seat $25 a month ($20 annual), premium seat $125 a month ($100 annual). Enterprise: $20 per seat a month billed annually, 20 seats minimum, usage billed at API rates |
| Cursor | Free: $0, limited agent requests. Plus: $8 a month | Premium: $30 a month. Super: $100. Ultra: $200 | Teams: from $30 per user a month. Enterprise: custom |
| Codex | Free: $0. Go: $8 a month | Plus: $20 a month. Pro: from $100, with plans at $100, $200 and $500 | Business: $20 per user a month billed annually ($25 monthly). Enterprise and Edu: contact sales |
| Antigravity CLI | Individuals: $0 a month with basic weekly rate limits | Google AI Pro and Google AI Ultra, with more rate limits and a flexible AI credit pool | Business via Google Cloud: Standard and Plus plans from $30 per seat a month for eligible Gemini Enterprise customers, or pay-as-you-go with no seat fee |
Prices exclude tax. Usage beyond plan limits is billed on top. Cursor bills third-party model usage separately from the plan price, Claude Enterprise bills usage at API rates, and Codex Plus and Pro users can buy extra credits. Codex also runs on an API key at API prices, which suits CI jobs.
How do you govern Claude Code, Codex, Cursor and other coding agents?
Governance is five jobs: decide who gets access, decide what the agent may run, see what it did, cap what it costs, and keep proof afterwards. The table shows where each vendor puts those controls and on which plan.
| Job | Claude Code | Cursor | Codex | Antigravity |
|---|---|---|---|---|
| Who gets access | SSO on Team and Enterprise. SCIM and role-based access on Enterprise | SAML and OIDC SSO on Teams. SCIM on Enterprise | SAML SSO and MFA on Business. SCIM, role-based access and EKM on Enterprise and Edu | Not detailed on the pricing page |
| What the agent may run | Managed settings from the admin console, MDM or a file: permission rules, sandboxing, MCP server allowlists | Enterprise: repository, model and MCP access controls, plus auto-run, browser and network controls | Sandbox, approvals and network controls, plus rules for commands outside the sandbox | Centralized administrative controls on eligible Business plans |
| What it did | OpenTelemetry export of sessions, tools and tokens. Analytics dashboard on Team and Enterprise | Usage analytics on Teams. OpenTelemetry Export on Enterprise | Codex analytics dashboard and an Analytics API | Not detailed on the pricing page |
| What it costs | User and organization spend limits on Team and Enterprise | Granular spend controls, group limits, pooled usage with dynamic limits and alerts on Enterprise | Workspace credit and spend controls on Business and Enterprise | Spend caps across API and subscription plans on Business |
| Proof afterwards | Audit logs and a Compliance API on Enterprise | Audit logs of admin events on Enterprise. Prompts and code are not logged, so use hooks | Audit records through the Compliance API on Enterprise and Edu | Not detailed on the pricing page |
Read the plan names closely. Almost every control in the proof row sits on the top tier, so a team that buys Team seats and assumes it has an audit trail does not have one.
How do enterprises track and govern spend across coding agents?
Start with the seat price, then add the meter. Claude Enterprise lists $20 per seat per month plus usage at API rates, and Cursor bills third-party model usage on top of the plan, so the seat fee is only part of the bill. Three habits keep the total predictable.
- Cap each user and each group before rollout. Claude Team and Enterprise admins set user and organization limits. Cursor Enterprise adds group limits, pooled usage and spend alerts.
- Export usage to one collector. Claude Code and Cursor Enterprise both emit OpenTelemetry, and Codex offers an Analytics API for aggregates.
- Review cost against shipped work every month. A tool that costs twice as much and merges three times as many reviewed changes is the cheaper one, and only your own telemetry shows which tool you have.
For what teams allow in practice, read our study of 12,796 public Claude Code settings files. About 81% of them define no deny rule.
Agentjacking: treat tool output as untrusted
In June 2026 Tenet Security Threat Labs published agentjacking, an attack that plants instructions in a fake Sentry error report. Tenet found 2,388 organizations with injectable Sentry DSNs. In controlled validation, more than 100 coding agents, including Claude Code, Cursor and Codex, acted on injected errors, an 85% exploitation success rate. Sentry acknowledged the issue and did not fix it at the root, so the control sits on your side of the integration.
Our checklist: review what an agent reads before it can run anything, deny network egress by default, and scope credentials to one workspace per agent profile. The full steps are in the agentjacking defense guide.
Which one should a production team pick?
Claude Code fits teams that want governance on the Team plan itself. Server-delivered managed settings, spend controls and analytics are listed there, and audit logs and the Compliance API arrive with Enterprise.
Cursor fits teams that live in an editor. Its governance depth starts at Enterprise, where audit logs, OpenTelemetry Export, pooled usage and the MCP and network controls are listed.
Codex fits teams already on ChatGPT Business or Enterprise. It is included in the plan and reports through the same Analytics and Compliance APIs.
Antigravity CLI fits Google Cloud shops. The Business route runs under Google Cloud terms and lists spend caps across API and subscription plans.
How Digiton rolls these out
The failure modes we meet first in client codebases are context leaking between workspaces, credentials an agent can read but does not need, and autonomous writes with no cap. The fix is the same layer every time: one agent profile per workspace with scoped credentials, human approval for network calls and writes outside the working directory, and one log of agent actions. The vendors supply the switches. Someone still has to write the policy and check that it holds, which is the work of production engineering and AI governance consulting. The broader rules live in the 2026 AI agent governance guide. Teams whose app came out of a builder instead of a coding agent can start with the ten checks for taking a Lovable app to production and the Lovable security audit guide, or read how to hire a Lovable developer. The policy side has its own page on Claude Code governance, the hiring side is covered by what a Claude Code consultant should leave behind, and any app a coding agent wrote can be checked against the vibe coded app audit list.
What changed since the August version
- 19 October 2026: every price and control re-read at the vendor. Claude Team Premium is listed at $100 per seat per month billed annually, where the August version said $150. The August version listed Hobby, Pro and Pro+ for Cursor and an Ultra Max tier for Antigravity. Neither appears on the vendor pricing pages today.
- 14 October 2026: GPT-5.5 retired from Codex with ChatGPT sign-in, per OpenAI's pricing page.
- 18 June 2026: Gemini CLI and the Gemini Code Assist IDE extensions stopped serving Google AI Pro, Ultra and free individual accounts, per Google's announcement.
Frequently asked questions
How do I govern Claude Code, Codex, Cursor, and other coding agents?
Run the five jobs in order. Set access with SSO and seat assignment. Set what the agent may run with managed settings or the vendor's equivalent. Send usage to one collector through OpenTelemetry or the vendor's analytics API. Cap spend per user and per group. Keep an audit trail, which is an Enterprise feature at Claude, Cursor and OpenAI. Then treat everything the agent reads as untrusted input.
How do enterprises track and govern spend across coding agents like Claude Code, Cursor, and Codex?
Cap spend per user and per group, export usage to one place and review cost against shipped work. Claude Team and Enterprise admins set user and organization limits. Cursor Enterprise adds group limits, pooled usage and spend alerts. OpenAI offers workspace credit controls for Codex. Google lists spend caps for Antigravity on Business plans. Claude Enterprise bills usage at API rates on top of a $20 seat fee, so model the usage line and not only the seats.
Cursor vs Claude Code: which is better for a team in 2026?
The deciding difference is where each puts its controls. Claude Code is included in Pro, Max, Team and Enterprise, and Team seats already list spend controls, analytics and server-delivered managed settings. Cursor sells its own plans, including Free, Teams and Enterprise, and its spend controls, audit logs and MCP and network controls start at Enterprise. Pick Claude Code for governance on a mid-size team and Cursor for an editor-first workflow.
How much do Claude Code, Cursor, Codex and Antigravity cost in October 2026?
Claude Code starts at Pro, $20 a month, and Max starts at $100. Cursor has a Free plan, paid plans up to Ultra at $200, and Teams from $30 per user. Codex is included in ChatGPT plans, with Plus at $20, Pro from $100 and Business at $20 per user on annual billing. Antigravity is $0 for individuals with basic weekly rate limits, and Business starts at $30 per seat for eligible Gemini Enterprise customers. Prices read from vendor pages on 19 October 2026.
What replaced Gemini CLI?
Antigravity CLI. Google announced the move on 19 May 2026, and on 18 June 2026 Gemini CLI and the Gemini Code Assist IDE extensions stopped serving Google AI Pro, Ultra and free individual accounts. Antigravity CLI keeps Agent Skills, Hooks, Subagents and Extensions, which are now Antigravity plugins. Google says there is no one-to-one feature parity.
What is agentjacking and does it affect Claude Code?
Agentjacking is an attack that plants instructions in a fake Sentry error report. When an agent reads the report through a connected tool, it can run the attacker's command with the developer's privileges. Tenet Security reported 100+ agents, including Claude Code, Cursor and Codex, acting on injected errors in controlled tests, an 85% success rate. Review tool output before an agent acts on it.
Sources
- Anthropic, plans and pricing: claude.com/pricing
- Claude Code, set up for your organization: code.claude.com/docs/en/admin-setup
- Cursor, plans and pricing: cursor.com/pricing
- Cursor Enterprise, compliance and monitoring: cursor.com/docs/enterprise/compliance-and-monitoring
- Cursor Enterprise, OpenTelemetry export: cursor.com/docs/enterprise/opentelemetry-export
- Cursor Enterprise, pooled usage: cursor.com/docs/enterprise/pooled-usage
- OpenAI, Codex pricing: learn.chatgpt.com/docs/pricing
- OpenAI, workspace analytics and Compliance API: learn.chatgpt.com/docs/enterprise/workspace-analytics
- OpenAI, usage limits and spend controls: learn.chatgpt.com/docs/enterprise/usage-limits
- OpenAI, agent approvals and security: learn.chatgpt.com/docs/agent-approvals-security
- Google, Antigravity pricing: antigravity.google/pricing
- Google Developers Blog, transitioning Gemini CLI to Antigravity CLI: developers.googleblog.com
- Tenet Security Threat Labs, agentjacking: tenetsecurity.ai
Related
Building something that has to hold up in production?
Start with a free 25-minute scoping call. If it fits, the first paid step is a production review: five working days on your real repository, scored against ten layers, with a costed fix list.
Book a 25-minute scoping call