On June 12, 2026, Anthropic confirmed that its agent products would move to a unified credit pool on June 15, 2026. The announcement, posted on the official Anthropic pricing page, ended the flat-rate access model for Claude Code, the Agent SDK, and console-based agent runs. Free users lost a set number of daily agent messages. Pro and Max subscribers gained a monthly credit allowance that pulls from the same pool as API calls. This is not a price cut. It is a structural change. The pool resets monthly, but unused credits do not roll over in most cases. Heavy users will hit the cap faster than they did under the old daily reset system.
The change hit three groups. Free-tier users saw agent access shrink to a smaller monthly credit pool, with automatic cutoffs after exhaustion. Pro subscribers at $20 per month got a fixed credit pool that replaced unlimited agent preview access. Max subscribers at $100 or $200 per month faced higher allowances but new overage fees. Developers using the Anthropic API now see agent tool calls deduct from the same credit balance as token generation. According to the official Anthropic changelog, the move was meant to stop abuse and align costs. It also shifted risk back to users, especially builders who ran long-horizon agents without tracking spend.
Why now matters. Anthropic had subsidized agent workloads since late 2025, when Claude Code and the Agent SDK launched with flat-rate access. That subsidy became expensive as agentic AI billing crisis reports showed a wave of free-tier users running multi-hour coding agents. Competitors were also moving. Google tightened Gemini API free access in June, a shift covered in Google Gemini API free tier tightened. Anthropic did not want to be the only major lab still offering unlimited flat-rate agent access. The credit pool aligned the company with a broader industry turn toward usage-based billing.
Users should not mistake this for a simple rate limit update. Credit pools are not the same as message caps. A single Claude Code run can consume thousands of credits because every tool call, file read, and code edit draws down the balance. Anthropic published a credit consumption table on its pricing page. It showed that one hour of medium agent activity on Claude Opus 4.8 could consume roughly 18,000 credits, while a lightweight Claude Haiku agent run used about 2,400 credits. That difference determines whether a free user burns through a 50,000-credit monthly pool in days or weeks. The new model rewards careful prompt and tool design and punishes open-ended loops.
How Do the Top Options Compare?
| Plan | Pre-June 15 Access | June 15 Credit Pool | Overage Policy |
|---|---|---|---|
| Free | Flat-rate daily agent messages, 5-hour reset | 50,000 credits per month | Hard cutoff, no rollover |
| Pro ($20/mo) | Unlimited agent preview | 500,000 credits per month | $5 per 100,000 extra credits |
| Max ($100/mo) | Unlimited agent preview with priority | 2,000,000 credits per month | $4 per 100,000 extra credits |
| Max ($200/mo) | Unlimited agent preview with priority | 5,000,000 credits per month | $3 per 100,000 extra credits |
| API developers | Separate token billing, no agent surcharge | Unified credit draw including tool calls | Standard API overage rates |
Credit values are based on Anthropic’s June 12, 2026 pricing page and may change. One credit does not equal one token. Tool calls, file edits, and multi-step reasoning consume additional credits. Free AI News may earn a commission if you sign up for a paid plan through links on this site. This does not affect our reporting.
1. Free Tier Agent Access , Casual users testing Claude Code with low volume
Free users experienced the sharpest change. Before June 15, the free tier included flat-rate agent messages with a five-hour reset window, documented in Claude free tier changes 2026. After June 15, that model disappeared. Anthropic replaced it with a 50,000-credit monthly pool shared across Claude Code, Agent SDK runs, and console agent tests. A single Opus 4.8 agent run could consume more than a third of that pool in under two hours.
The lower-cost Haiku path helped. Anthropic said Claude Haiku 4.5 agent runs drew roughly 2,400 credits per hour of moderate tool use. That means a free user could stretch 50,000 credits to about 20 hours of light agent work. But heavy use with Opus or Sonnet burned through the allowance in days. The monthly reset offered no daily recovery valve, and users had to wait until the next billing cycle to continue.
Who suffers most? Students and indie developers who relied on free Claude Code for small projects and learning now face a hard stop. The credit pool also resets monthly, not daily, so a bad week cannot be fixed by waiting a day. This is a significant retreat from Anthropic’s earlier free-tier generosity.
Key strengths:
- ✅ 50,000 free monthly credits still allows light testing
- ✅ Haiku 4.5 lowers credit burn for simple tasks
- ✅ No credit card required for free users
- ✅ Clear dashboard shows remaining balance
- ❌ No rollover means wasted credits at month end
- ❌ One Opus run can consume over 18,000 credits
- ❌ Hard cutoff halts work until next cycle
Who it’s for: Casual testers who need only a few hours of free agent use per month.
2. Pro Plan Agent Credit Pool , Individual developers who need predictable monthly access
Pro subscribers saw the end of unlimited agent preview access. The $20 monthly plan now includes 500,000 credits, which at first sounds generous. The official Anthropic pricing page showed that 500,000 credits equals roughly 208 hours of Claude Haiku 4.5 agent use or 27 hours of Claude Opus 4.8 agent use. The exact split depends on tool-call density, context length, and model choice.
This is the most controversial tier. Pro users had become accustomed to flat-rate agent access during the beta period. The new pool forces them to choose between cheap Haiku runs and expensive Opus runs. Many developers told Free AI News they would shift to Haiku by default. The credit pool also unified API and console usage for Pro, which means a heavy weekend of agent coding can leave no credits for weekday API experiments.
Anthropic framed the move as a subsidy correction. The Anthropic ends agent subsidy June 15 credit pool replaces flat-rate access post said flat-rate Pro access was never designed for production-scale agent workloads. But the result is still a price increase for anyone who used Claude Code more than a few hours per week. Overage fees are $5 per 100,000 extra credits.
Key strengths:
- ✅ 500,000 credits covers moderate weekly agent use
- ✅ Haiku runs are cheap, around 2,400 credits per hour
- ✅ Same $20 base price for now
- ✅ Unified credit pool simplifies billing
- ❌ Unlimited agent preview access is gone
- ❌ Opus 4.8 burns credits 7.5 times faster than Haiku
- ❌ Overage fees add up quickly during long coding sessions
Who it’s for: Individual developers who want predictable monthly costs and can stay within 500,000 credits.
3. Max Plan Agent Credit Pool , Heavy users and small teams with high-volume agent workloads
Max subscribers got larger pools but no escape from the new credit system. The $100 per month Max plan includes 2,000,000 credits, while the $200 per month Max plan includes 5,000,000 credits. Anthropic set overage fees at $4 and $3 per 100,000 credits respectively, a lower rate than Pro. This is the clearest signal that Anthropic wants high-volume users to move to the top tier.
Even Max is not unlimited. A small team running multiple Opus 4.8 agents for eight hours a day could hit 2,000,000 credits within a week. Anthropic’s own consumption examples showed that a single heavy coding agent with frequent file edits and long context can consume 60,000 credits per hour on Opus 4.8. At that rate, the $100 Max plan covers about 33 hours of Opus agent time per month.
The credit shift aligns Max with Anthropic’s enterprise pricing direction. The company has been pushing usage-based contracts since early 2026. For Max users, the benefit is lower overage pricing and priority access. The cost is the end of unlimited previews.
Key strengths:
- ✅ Larger credit pools: 2M or 5M per month
- ✅ Lower overage rates than Pro
- ✅ Priority access during peak hours
- ✅ Haiku runs stretch the pool much further
- ❌ Unlimited preview access removed for all paid tiers
- ❌ Heavy Opus use can exhaust 2M credits in under two weeks
- ❌ No unlimited option even at $200 per month
Who it’s for: High-volume developers and small teams that can justify $100 to $200 monthly spend.
4. API and Console Developers , Builders who need granular control over agent costs
API developers faced the most technical shift. Before June 15, agent tool calls and subagent runs were billed as token usage with no separate agent surcharge. After June 15, every agent action pulls from a single credit pool. Anthropic published a conversion table on its official Anthropic pricing page showing one credit equals roughly one token for text, but tool calls, file operations, and reasoning steps add fixed credit costs. A single tool call can cost 150 to 1,200 credits depending on context size.
This change broke many developer assumptions. Budget forecasts based on token counts no longer hold. An agent that reads 500 lines of code and makes 30 edits might consume 80,000 credits, even if the raw token count suggested only 30,000. Developers who built dashboards around Anthropic’s old billing must update them. The unified pool also means API and console usage now compete for the same balance.
The independent analysis from Stanford HAI’s AI Index noted that vendor billing complexity was a top barrier for developer adoption. This shift adds complexity. But it also exposes true agent costs. Developers who instrument their agents can now see exactly how much tool calls cost and optimize accordingly. That may be the only silver lining in an otherwise more expensive model.
Key strengths:
- ✅ Granular credit table shows tool-call costs
- ✅ Haiku and Sonnet models keep agent runs cheap
- ✅ Unified billing across API and console
- ✅ Better for cost attribution in production
- ❌ Existing token-based budgets are obsolete
- ❌ Tool calls add unpredictable credit overhead
- ❌ API and console usage compete for one balance
Who it’s for: Developers who can instrument, monitor, and optimize agent credit consumption.
5. Claude Code Power Users , Developers who live in the terminal and run long sessions
Claude Code users got a double hit. The credit pool replaced flat-rate access, and Anthropic removed the five-hour reset that previously allowed users to recover. This was covered in Claude Code limits jump 50 paid vs free 2026. Paid users now have higher credit ceilings but no unlimited fallback. Free users must ration carefully because a long debugging session can consume an entire day’s worth of credits in one go.
The shift is especially painful for users who run agentic coding loops. Unlike a simple prompt, an agent loop can call the model dozens of times, read files, run tests, and revise code. Anthropic’s examples showed that a standard Claude Code session using Claude Sonnet 4.5 can consume around 12,000 credits per hour. A power user running four hours per day would need about 1.44 million credits per month. That exceeds the Pro plan completely and pushes users toward Max.
For some users, this will mean downgrading model choices or switching tools. Anthropic’s credit pool may be more transparent than hidden multipliers, but it is not cheaper for power users. Free AI News may earn a commission if you sign up for a paid plan through links on this site. This does not affect our reporting.
Key strengths:
- ✅ Higher credit ceilings for paid users
- ✅ Terminal usage still supported with clear accounting
- ✅ Opus, Sonnet, and Haiku options let users control cost
- ❌ Five-hour reset removed for free users
- ❌ Long agent loops burn credits quickly
- ❌ Pro plan insufficient for daily power use
Who it’s for: Terminal-native developers who can track credit burn and adjust models.
Frequently Asked Questions
What exactly changed on June 15, 2026?
Anthropic replaced flat-rate agent access with a unified monthly credit pool. Free, Pro, Max, and API users now draw from one balance that resets monthly. Unused credits do not roll over.
How many credits do free users get?
Free-tier users received 50,000 credits per month. A single heavy Claude Opus 4.8 agent run can consume 18,000 credits, so the pool can disappear within days.
Does the Pro plan still include unlimited agent access?
No. Pro now includes 500,000 credits per month. Overage fees are $5 per 100,000 credits. Unlimited agent preview access ended on June 15.
Are Max plans affected?
Yes. The $100 Max plan includes 2,000,000 credits, and the $200 Max plan includes 5,000,000 credits. Both lost unlimited previews but have lower overage rates.
Do credits roll over month to month?
No. Anthropic’s June 12 announcement stated unused credits do not roll over. The pool resets on the billing date.
Where can I see the official credit consumption table?
The official Anthropic pricing page has the credit table. It shows per-hour estimates for Haiku, Sonnet, and Opus agent runs.
What Should You Remember?
- Credit pool: Anthropic ended flat-rate agent access on June 15, 2026 for all tiers.
- Free tier: 50,000 monthly credits replace the old five-hour reset for free users.
- Pro plan: $20 Pro now includes 500,000 credits with $5 per 100,000 overage.
- Max plans: $100 Max gets 2,000,000 credits; $200 Max gets 5,000,000 credits.
- API billing: Tool calls and subagents now draw from the same credit pool as tokens.
- No rollover: Unused credits reset monthly and cannot be carried forward.
Free AI News is an independent editorial publication. Information about AI pricing, free-tier limits, and features changes frequently and may become outdated. Always verify current details through the vendor’s official pages. Affiliate links may earn a commission at no cost to you, and never affect our reporting.
