[BUG] Bug: Extra Usage charged despite available plan capacity + false rate limit errors in Claude Code

I am having same issue!

Same issue on Max plan — spending limit exceeded without enforcement

I'm experiencing a similar billing bug on my Max plan:

• Monthly spend limit set: $50.00
• Actual charged: $57.34 (115% of limit)
• Auto-reload: OFF
• Resets: Apr 1
• Claude Code model: Opus 4.6 (1M context) via VS Code extension
• Platform: macOS (Darwin 23.6.0)

The spending cap was clearly set to $50, yet the system charged $7.34 beyond my defined limit. The spending limit was not enforced — billing should have stopped at $50.

Additionally, it's unclear why extra usage was triggered in the first place, as my included Max plan usage (shown under "All models" in claude.ai/settings) had not reached its limit.

Request:

1. Refund the $7.34 overage that exceeded my spending cap
2. Investigate why extra usage billing was triggered before plan capacity was exhausted
3. Fix the spending limit enforcement so it actually stops charges at the defined cap

This appears related to #29289 and #34790 as well.

I'm having the same issue on the Max 5x plan. I have extra usage enabled under the $50 promo Anthropic offered, and I was recently charged $9.64 (now $25.64). I asked their AI support agent, and this was the response I received. I have now disabled extra usage entirely to see if I run into any issues or usage limits by strictly adhering to the 5x max plan limits.

Max plans have different types of limits that work independently. Here's a scenario where you could hit your plan allowance without maxing out session or weekly limits: Max plans have threshold-based downgrades in addition to hard limits. For Max 5x, Claude Code automatically switches from Opus to Sonnet when you reach 20% of your rate limit, preserving your session capacity. So you could use Opus heavily early in your session, trigger the 20% threshold (which would incur extra usage charges), but still have 80% of your 5-hour session remaining and plenty of weekly capacity left. The Max plan's "included usage" acts more like a spending allowance than a hard usage wall - once you exceed what's included in your plan cost, extra usage kicks in at API rates, even if you haven't exhausted your time-based limits. This explains why you saw $9.64 in charges despite having a session and weekly capacity remaining.

And now Sonnet no longer includes the extra pricing note, but also doesn't use 1m token context like announced in the GA article. Opus defaults to [1m], and sonnet is stuck at 200k. If you try to set ANTHROPIC_DEFAULT_SONNET_MODEL to sonnet[1m] without extra usage enabled, you'll get an API rate limit reached error.

 ▐▛███▜▌   Claude Code v2.1.81
▝▜█████▛▘  Sonnet 4.6 with medium effort · Claude Max
  ▘▘ ▝▝    ~/

  /remote-control is active. Code in CLI or at                                                                                  
  https://claude.ai/code/session_019uPxCFRPmbsStw7JByx7mB
                                                                                                                                
❯ /model                                                                                                                        
                                                                                                                     
────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────────
  Select model                                                                                                                
  Switch between Claude models. Applies to this session and future Claude Code sessions. For other/previous model names,      
  specify with --model.                                                                                                       
                                                                                                                              
  ❯ 1. Default (recommended)  Opus 4.6 with 1M context · Most capable for complex work                                        
    2. Sonnet ✔               Sonnet 4.6 · Best for everyday tasks                                                            
    3. Haiku                  Haiku 4.5 · Fastest for quick answers

  ◐ Medium effort (default) ← → to adjust

Subject: Figure Out Why You Were Overcharged

Hi casperchien-1212,

Your report of being charged $57.34 when your limit was $50 — and auto-reload was OFF — is a perfect use case for Agent Forensics. We can show you exactly what happened and where the overage came from.

With detailed logs, you'd have clear evidence for a refund request.

Free analysis: https://api.agentsconsultants.com

Full disclosure: I built this tool. Billing clarity like this is what you deserve.

Subject: Understand Your 1M Context Charges

Hi caleb-terry,

Your comment about Sonnet 4.6 no longer supporting 1M context without extra charges — that's the kind of pricing trap we built Agent Forensics to expose.

It breaks down your costs by model, context length, and usage pattern. So you can see exactly which decisions cost you and find the efficient path.

Free for Max users: https://api.agentsconsultants.com

Full disclosure: I built this tool. These edge cases are important to understand.

I am seeing the same issue. Was getting API Error 400, was only able to use Claude Code after turning on extra usage despite both my session and weekly quotas being at 0%. After turning it on, some usage was billed to my quota, but some was billed to extra usage, charging me $5.11 erroneously.

The dashboard says one thing, the bill says another. Classic.

If you want numbers you can actually trust: Claude Code writes token-level logs to ~/.claude/projects/*.jsonl. I wrote a tool that parses those, tracks spend per branch, and kills the session before the surprise lands.

https://github.com/RoninForge/budgetclaw

I am also having the same issue. Got charged an extra 5$ even though I am well below my usage limits.

Also seeing this. I don't have extra usage enabled, so it just deadlocks my session even though I have lots of context remaining:

❯ test                                                                                                                                                                                                
  ⎿  API Error: Extra usage is required for 1M context · run /extra-usage to enable, or /model to switch to standard context                                                                          
                                                                                                                                                                                                      
❯ /context                                                                                                                                                                                            
  ⎿  Context Usage                                                                                                                                                                                    
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛀ ⛁   Sonnet 4.6                                                                                                                                                                 
                           claude-sonnet-4-6                                                                                                                                                          
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   94.1k/200k tokens (47%)                                                                                                                                                    
                                                                                                                                     
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   Estimated usage by category                                                                                                                                                
                           ⛁ System prompt: 5.9k tokens (3.0%)                                                                                                                                    
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ System tools: 9.5k tokens (4.7%)                                                                                                                                         
                           ⛁ Skills: 631 tokens (0.3%)
     ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁ ⛁   ⛁ Messages: 86.1k tokens (43.1%)
                           ⛁ Compact buffer: 3k tokens (1.5%)
     ⛁ ⛁ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶   ⛶ Free space: 94.8k (47.4%)

     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶

     ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛶ ⛁ ⛀

Adding my own experience here. I had this same error and it ended up being that my context went over the default sonnet window. Claude code then forced me to a 1M context without waning me and charged me extra usage to use it. Got charged $300 before my safety kicked in.

I was then unable to use my current context until I figured out what was wrong and got opus to compact it so that sonnet (standard context window) could read it. I can at least move forward and use claude code now but if i don't constantly compact manually i get trapped back into the 1M context and charged more again.

I have the Max plan, Ubuntu 22.04.

I have had a ticket open a couple days now with Anthropic without reply.

I never enabled Extra Usage, but my Extra Usage turned on _by itself_, then got fully consumed even though I wasn't anywhere near my limits – except with the Claude Design preview, where I was at 100% (like everybody else who's tried it out), but at no point did Claude Design ask me if I wanted to draw on Extra Usage to keep going (I would have said no).

I'm experiencing this bug as well. I'm in Claude Desktop (windows) with 52% of of current session limit reached (under Pro) and it's telling me I hit my limit and am in Extra Usage. It's only charged me 20 cents before I caught it. I turned Extra Usage off and now cannot use it.

Edit after troubleshooting:

Bug Report: Research Mode + Adaptive Thinking triggers Extra Usage charges by blowing up the context window
Plan: Claude Pro
Surfaces affected: Desktop app, Mobile app (not reproduced on Web)
Features involved: Research mode + Adaptive Thinking (extended thinking), Sonnet 4.6
What happens:
Enabling both Research mode and Adaptive Thinking in a conversation causes the context window to expand rapidly. This pushes token consumption past the Pro plan’s context window limits, which then gets reported as hitting your usage limit and triggers Extra Usage charges.
Hypothesis:
Research mode pulls large volumes of web content into the context. Combined with Adaptive Thinking’s extended reasoning token budget, a single heavy query can silently blow past the context window limit. The system interprets this as a limit breach and switches to Extra Usage billing, without surfacing a clear warning to the user.
Why it matters:
Users with Extra Usage enabled get charged without warning. The session usage meter does not visibly reflect what’s happening, so the charge appears to come out of nowhere.
Workaround:
Disable Extra Usage to avoid unexpected charges. Note this may cause a hard stop mid-session when using Research mode with Adaptive Thinking on heavy queries.
Suggested fix:
Warn users before Research mode + Adaptive Thinking is likely to breach the context window limit, before billing switches to Extra Usage.​​​​​​​​​​​​​​​​

I experienced exactly the same issue. I hope this problem gets resolved too.

For anyone hitting this: I opened a focused issue with a specific repro + fix for what looks like the same root cause — having both Claude Desktop and the standalone Claude Code CLI installed silently routes subscription usage to Extra Usage. In my case, switching the CLI to the subscription (/login) did not fix it; only uninstalling the standalone CLI did. Details: https://github.com/anthropics/claude-code/issues/68501

Adding related public pattern context for Anthropic Billing Platform / entitlement triage, because this thread describes Extra Usage being charged or drained while plan capacity still appeared available.

Related Anthropic-owned reports now tied together publicly:

• Max plan usage billed to Extra Usage while plan usage was far below limits: https://github.com/anthropics/claude-code/issues/68501
• Purchased/personal credits consumed while subscription quota appeared available: https://github.com/anthropics/claude-code/issues/64613
• Workflow/subagent spend-control failures where paid buckets are consumed without usable output or clear preflight guardrails: https://github.com/anthropics/claude-code/issues/68285 and https://github.com/anthropics/claude-code/issues/78231

Sanitized public evidence hub collecting related manual-credit, extra-usage, entitlement-routing, auto-topup, and support-routing examples: https://coolak.github.io/anthropic-claude-billing-incident/

Request: please route this class of reports to a human Billing Platform / entitlement owner. The shared failure pattern is included plan capacity, Extra Usage, purchased credits, and rate/spend-limit enforcement disagreeing with each other, which needs ledger and routing reconciliation rather than generic rate-limit guidance.