On May 13, 2026, OpenAI offered two months of free Codex to enterprises willing to switch from Claude within thirty days. The promotion, posted by Sam Altman on X, requires manual application review rather than instant activation. Anthropic responded the same day with a 50%1 weekly quota lift on Claude Code through July 13 for Pro, Max, Team, and Enterprise plans.
The May 13 Exchange: What Each Side Announced
The sequence moved fast. Ramp published its AI Index2 showing Anthropic ahead of OpenAI in U.S. business adoption for the first time. Within roughly an hour, Altman posted the Codex offer on X. About 45 minutes after that, Anthropic announced the 50%1 weekly limit increase at 2:19 PM PT, according to a timeline posted on LinkedIn3.
Altman’s offer is not automatic. Enterprises must submit an application through a promotional form and await manual review, according to coverage of the announcement4. Anthropic’s counter was broader but temporary: the quota lift runs through July 13, 2026, and applies to Pro, Max, Team, and seat-based Enterprise plans, per Anthropic’s announcement1.
Ramp Index: Why Anthropic’s Overtake Triggered the Response
The trigger was Ramp’s May 2026 AI Index2, published May 13. The report found Anthropic at 34.4%2 of U.S. business buyers versus OpenAI at 32.3%2, marking the first time Anthropic led the category.
The ranking shift is what turned a routine product update cycle into a public pricing skirmish. OpenAI had already removed the fixed seat fee for Codex’s business tier, according to one analysis5, so adding two free months was a natural escalation. Anthropic had to respond with something that did not require a structural pricing change, and a temporary quota lift fit the constraint.
The Pricing Chessboard: Seat Fees, Token Billing, and Third-Party Metering
The current pricing structures diverge in ways that matter for total cost of ownership. OpenAI removed the fixed seat fee5 for Codex’s business tier ahead of this promotion. Anthropic, by contrast, briefly tested removing Claude Code from the $206 monthly Pro tier6 for new users around April 22, 2026 — a change that lasted roughly twelve hours before being rolled back. [Updated June 2026] Anthropic’s head of growth confirmed the removal was a limited A/B test on approximately 2% of new signups; existing Pro and Max subscribers were not affected, and the pricing page was restored. The episode nonetheless signaled Anthropic’s intent to price Claude Code separately from its base tier.
OpenAI’s promotion specifically targets enterprises affected by Anthropic’s announced plan to charge separately for third-party agent tool usage, according to the same report5. [Updated June 2026] Anthropic subsequently paused that billing split before it took effect on June 15, 2026, and announced it would give advance notice before shipping a revised version. The metering proposal remains on the table; the unit-economics risk it creates for teams running Claude Code with external tools has not been permanently resolved.
Anthropic has been preparing for this capacity race. On May 6, 2026, it permanently doubled Claude Code’s five-hour rate limits7 and removed peak-hours throttling for Pro and Max accounts, backed by a compute deal with SpaceX for more than 300 megawatts7 of capacity.
What the 60-Day Window Means for Procurement Teams
Two months of free Codex sounds like a straightforward trial, but the real math involves switching costs. Teams running Claude Code with custom tool integrations face migration work that a 60-day window may not cover. The offer also expires in roughly 30 days from announcement, meaning the application window closes around mid-June.
Anthropic’s quota lift, meanwhile, expires July 13. Procurement teams evaluating both offers must decide whether temporary capacity relief outweighs the structural shift in Anthropic’s pricing. The Max tier requirement for new Claude Code users means greenfield adoption is already more expensive than it was a month ago, while OpenAI is effectively paying enterprises to try Codex at zero upfront cost.
The Lock-In Risk Both Vendors Are Downplaying
Neither vendor is talking much about what happens after the promotional period ends. OpenAI wants workflow data and habit formation; two months is enough time for teams to build Codex-dependent processes. Anthropic wants to retain seats while it figures out how to monetize third-party agent usage without pushing buyers to competitors.
The structural question is whether flat-fee removal or per-token metering wins in enterprise budgets. OpenAI is betting that removing seat fees lowers the barrier enough to offset switching friction. Anthropic is betting that quota relief keeps existing users from leaving while it builds out capacity. Both assumptions rest on the same premise: that LLM tooling stickiness is high enough to survive a pricing war. For teams already integrated, it probably is. For teams still evaluating, May 13 made the decision more complicated than a feature comparison.
What Has Changed Since May 13
[Updated June 2026] The competitive landscape shifted again in the weeks following the initial exchange, in ways that reduce switching friction on the OpenAI side and complicate the cost calculus on Anthropic’s.
One-click migration tool. OpenAI paired the two-month promo with a migration tool built into the Codex desktop app. Opening Settings and selecting “Import other agent setup” triggers a scan of the local machine for Claude Code configurations — skills, hooks, MCP servers, subagent definitions, instruction files, and up to 30 days of session history — and migrates what it can automatically. Items that cannot be migrated automatically open a follow-up thread. The practical effect is that the per-developer switching cost dropped from days of manual config work to an afternoon. For enterprises evaluating during the promo window, the migration tool addresses the most common objection: that two months is not enough time to complete migration alongside normal work.
Codex 0.137.0 (June 4, 2026). OpenAI shipped multi-agent v2 with per-thread runtime persistence, closing the session-continuity gap that made agent-based workflows on Codex fragile relative to Claude Code. The same release added enterprise monthly credit controls visible in the status bar and cloud-managed configuration bundles for EDU workspaces. A prior release added session archiving and command-safety hardening that blocks repo-provided Git hooks from running during /diff reviews — a targeted fix for a supply-chain attack vector in malicious repositories. A full breakdown of what these releases change for switching teams is in OpenAI Upgrades Codex Right as Teams Weigh Leaving Claude Code.
Anthropic paused the Agent SDK billing split. Anthropic announced a June 15, 2026 launch for a separate credit pool that would have moved Agent SDK, claude -p, GitHub Actions, and third-party tool calls off subscription limits onto metered credits ($20 for Pro, $100 for Max 5x, $200 for Max 20x, no rollover). It paused the change before it took effect and said it would give advance notice before shipping a revised version. The pause removes the hard deadline that was driving urgency for enterprise procurement teams — but the underlying intent to meter agentic workloads separately has not been withdrawn. Teams that sized their Codex evaluation against the June 15 deadline may need to reset their comparison timeline.
Anthropic’s third-party tool access reinstated under metered model. Third-party tools that authenticate through Claude subscriptions — including OpenClaw, Zed, and Conductor — were initially banned, then reinstated. They are currently allowed but will move to the metered credit pool whenever Anthropic ships the revised billing change. For enterprises choosing between vendors partly on ecosystem breadth, this is a variable rather than a settled line item. The full policy history — ban, reinstatement, and credit-pool design — is covered in Anthropic Bans Third-Party Subscription Auth: The Three-Stage Repricing.
Frequently Asked Questions
How should teams budget for Anthropic’s third-party agent tool metering?
Anthropic’s separate billing for third-party agent tool usage shifts costs from a predictable per-seat line to a variable expense that scales with how often Claude Code calls external tools, APIs, databases, shell commands. Teams running agentic workflows with heavy tool invocation could see metering charges exceed the $100/month Max tier base price by a significant margin. Procurement teams should instrument tool-call volume in their current Claude Code setup before projecting total cost under either vendor.
How does the 50% quota lift stack on top of Anthropic’s earlier capacity upgrade?
Anthropic’s May 6 SpaceX deal permanently doubled Claude Code’s five-hour rate limits and removed peak-hours throttling. The 50% weekly quota lift announced May 13 stacks on top of those already-doubled limits, meaning current Pro and Max users are effectively operating at roughly 3x their pre-May throughput. OpenAI’s two free months, by contrast, provide a temporary price break with no underlying capacity improvement, the offer expires regardless of usage volume.
Can workflow data or agent context transfer between Codex and Claude Code?
[Updated June 2026] OpenAI has since shipped a built-in migration tool in the Codex desktop app that imports Claude Code skills, hooks, MCP servers, subagent definitions, instruction files, and up to 30 days of session history. What it cannot port automatically is surfaced in a guided follow-up thread. Anthropic has not announced equivalent migration tooling or data export guarantees. Two months of Codex usage builds institutional knowledge — trained habits, tuned subagents, integrated workflows — that does not port cleanly back to Claude Code. This asymmetry favors OpenAI: the longer a team operates on Codex, the costlier a hypothetical switch-back becomes, even after the promotional period reverts to standard billing.
Does the Ramp Index lead reflect true enterprise-grade adoption?
Ramp’s dataset is built from corporate-card transaction data, which skews toward startups and SMBs and misses Azure-OpenAI deals negotiated through Microsoft enterprise agreements. Large enterprises running OpenAI models via Azure likely do not appear in the 32.3% figure, meaning OpenAI’s actual business footprint may be understated. Procurement teams should treat the index as a directional signal for startup-heavy segments rather than a complete market map.
