🧭 Programmatic Usage Gets Its Own Monthly Credit Pool — What Every Claude Developer Needs to Do Before June 15

Starting June 15, 2026, Anthropic is separating how programmatic and interactive usage are billed across all Claude subscription tiers. Calls made via the Agent SDK, claude -p (non-interactive mode), Claude Code GitHub Actions, and third-party apps authenticating through the Agent SDK will no longer draw from the same pool as Claude.ai chat or terminal-interactive Claude Code sessions. Instead, they move to a dedicated monthly credit pool billed at full API rates — and you must actively claim those credits.

Credit amounts by plan

• Pro ($20/mo): $20 in monthly programmatic credits
• Max 5× ($100/mo): $100 in monthly programmatic credits
• Max 20× ($200/mo): $200 in monthly programmatic credits
• Team and Enterprise: credits scale with seat count and plan — check your billing dashboard for your allocation

Credits do not roll over — unused programmatic credits expire at the end of each billing month. Anthropic will send a claim link to subscribers around June 8; if you miss the claim window, you can still activate credits from your account settings page.

What this means for Claude Code and Agent SDK pipelines

Practically, this change affects two groups differently:

• Interactive Claude Code users (terminal sessions, slash commands, in-session completions): no change. These remain on your standard subscription limit.
• Automated pipelines (GitHub Actions, cron-scheduled Claude Code runs, programmatic Agent SDK calls, MCP server tool calls initiated outside a live terminal session): these will now consume the programmatic credit pool. Once that pool is exhausted, automated calls fail — they do not fall back to your interactive limit.

The OpenClaw / third-party agent reversal

This change also reverses Anthropic's April 2026 ban on third-party agents accessing Claude via subscription credentials. Apps like OpenClaw and other Agent SDK integrations are now permitted again — but metered against the programmatic credit pool rather than the shared subscription limit. The April ban was a temporary measure while Anthropic designed this metered architecture.

# Check if your workflow is interactive or programmatic
# Programmatic = anything that passes -p / --print or runs headlessly

# This runs interactively (uses subscription limit):
claude "explain this function"

# This runs programmatically (will use credit pool from June 15):
claude -p "explain this function"
claude --print "explain this function"

# GitHub Actions / cron: always programmatic
# MCP tool calls from automated servers: programmatic

🧭 Code with Claude Heads to Tokyo — Anthropic's First Developer Conference in Asia (June 10–11)

Anthropic's Code with Claude conference series completes its 2026 world tour with a Tokyo stop on June 10–11, making it Anthropic's first developer conference on Asian soil. The format mirrors San Francisco (May 6) and London (May 19): a full-day keynote plus hands-on workshops on June 10, followed by an Extended day on June 11 reserved for independent developers and early-stage founders. Both days are livestreamed in English and Japanese with simultaneous interpretation.

What to expect

• Product announcements: The SF and London events produced major unveilings — multi-agent orchestration, Claude Outcomes, Dreaming — and the Tokyo keynote is expected to continue that cadence. Anthropic has previewed at least one "Japan-first" announcement tied to its growing regional partnerships.
• Workshop tracks: Sessions cover Claude Code automation, MCP server development, enterprise deployment patterns, and agentic pipeline design. The Extended day (June 11) adds a founder track specifically for pre-Series-A teams.
• Simultaneous interpretation: All main-stage sessions run in both English and Japanese, removing the language barrier that historically made Western AI conferences inaccessible to a large portion of Japan's developer community.
• Livestream: If you cannot attend in person, register for the free livestream via the official event page — sessions are also archived on demand after the event.

🧭 Claude Sonnet 4 and Opus 4 Base Models Retire June 15 — Migration Guide

Anthropic has confirmed that Claude Sonnet 4 (the original launch variant, not Sonnet 4.6) and Claude Opus 4 (the original, not Opus 4.7 or 4.8) will be retired from the API on June 15, 2026 — the same date as the programmatic credit pool change. API calls targeting these model IDs after June 15 will return a deprecation error. The path forward is straightforward but requires updating model strings before the cutover date.

Which model IDs are affected

• Retiring: claude-sonnet-4-20250514, claude-sonnet-4-latest (when pinned to the original release)
• Retiring: claude-opus-4-20250514, claude-opus-4-latest (original launch build)
• Not affected: claude-sonnet-4-6, claude-sonnet-4-7, claude-opus-4-6, claude-opus-4-7, claude-opus-4-8 — all remain fully available

Migration steps

1. Audit your model strings. Search your codebase for claude-sonnet-4-20250514, claude-opus-4-20250514, and any -4-latest aliases that could resolve to the original builds. A simple grep covers most cases:

   grep -r "claude-sonnet-4\|claude-opus-4" . \
     --include="*.py" --include="*.ts" --include="*.js" \
     --include="*.yaml" --include="*.env" \
     | grep -v "4-6\|4-7\|4-8"
   

2. Replace with the recommended successor. For most workloads, the direct upgrade is:
   • Sonnet 4 → claude-sonnet-4-6-20260101 (or claude-sonnet-4-7 for the latest Sonnet)
   • Opus 4 → claude-opus-4-8-20260529 (latest Opus with the strongest SWE-Bench and honesty scores)
3. Run your eval suite against the new model IDs before promoting to production. The honesty and context-handling improvements in 4.6+ mean some prompts that relied on the original model silently accepting flawed premises may now receive explicit corrections.
4. Update Bedrock/Vertex ARNs. If you're on managed cloud, cross-region inference profiles pointing to -4-latest may automatically resolve to a newer build, but explicitly versioned ARNs will break. Verify your ARN inventory matches the new model IDs.