Opus 4.6 (1M context) hits 'API Error: Rate limit reached' on Max subscription

Update: Sonnet 4.6 with 1M context has also stopped working now — same API Error: Rate limit reached error.

So the issue affects all models with 1M context, not just Opus.

• Opus 4.6 + 1M context — rate limited
• Sonnet 4.6 + 1M context — rate limited (was working earlier, just broke)
• Both models without 1M context — still work fine

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/27535
2. https://github.com/anthropics/claude-code/issues/27179
3. https://github.com/anthropics/claude-code/issues/26496

This issue will be automatically closed as a duplicate in 3 days.

• If your issue is a duplicate, please close it and 👍 the existing issue instead
• To prevent auto-closure, add a comment or 👎 this comment

🤖 Generated with Claude Code

I also noticed a separate but potentially related issue this morning — billing charges despite being well within my usage limits.

I have the extra usage limit set, and both my weekly and 5-hour rolling window were at essentially 0% remaining when the session started. Yet Claude Code somehow began charging credits for extra usage anyway. I only caught it because I happened to check my balance.

I got lucky — my limit was set conservatively, so only ~$2 was charged. But if I'd had a higher cap, I could have been hit with a much larger unexpected charge without noticing.

Expected behavior: No extra usage charges should occur when you're within (or at the bottom of) your usage windows.
Actual behavior: Credits were deducted as "extra usage" even though limits weren't exhausted.

@claude add also billing label

I am experiencing the exact same issue on the Claude Max ($200/mo) plan. Opus 4.6 with 1M context returns an immediate 'Rate limit reached' error, while other models work fine.

Hey, did some digging into this and I think I figured out what's actually happening here.

The "Rate limit reached" error seems to kick in specifically when your subscription usage is exhausted and your context has already grown past ~200k tokens. Here's the deal with how billing works for 1M context models:

• Up to ~200k tokens of context — this is covered by your regular Max subscription
• Anything above 200k tokens — gets billed as extra usage (the credits/pay-as-you-go stuff)

So the rate limit error isn't really a "rate limit" in the traditional sense — it's more like you've burned through your subscription quota and the overflow portion (above 200k) needs extra usage credits to continue.

What this means in practice:

• If your session is already past 200k tokens and you're out of subscription usage → you need to have extra usage credits loaded to keep going
• If your session is under 200k tokens and you're hitting this → just switch to the regular (non-1M) Opus or Sonnet. The standard context window models are covered normally by your subscription

Basically the 1M context window is a bit of a trap if you don't realize the 200k+ portion eats into extra usage. Hope this helps someone save some debugging time 😅

@in0vik let me clarify what you're saying: if the 1m context _previously worked_ and stopped working with the symptoms of:

• api rate limit reached
• /models no longer displaying the 1m options

Even in the context of the usage panel showing plenty of remaining usage:

<img width="950" height="447" alt="Image" src="https://github.com/user-attachments/assets/4ef86ff5-08e3-4049-906b-28793602e8d3" />

Does this mean that there was likely a _different bucket_ of additional credits that are _not represented_ in the usage UI which the account was feeding off of? And that the behavior change (api limit reached, 1m models missing) indicates that this _other_ bucket has been disabled or depleted?

This might not be debug-able if you are not also the admin of your account.

exactly same issue, in the highest Max plane but when try to use 1m mode of opus tells me API Error: Rate limit reached error.
undergrade to normal opus works fine.

<img width="2412" height="1780" alt="Image" src="https://github.com/user-attachments/assets/a0f07f8a-2abf-45d2-ae89-59079bc8f08e" />

@in0vik let me clarify what you're saying: if the 1m context _previously worked_ and stopped working with the symptoms of:让我澄清一下你的意思：如果 1m 上下文_之前正常工作 _，后来出现以下症状停止工作： api rate limit reachedAPI 速率限制已达上限 /models no longer displaying the 1m options模型不再显示1米选项 Even in the context of the usage panel showing plenty of remaining usage:即使在使用情况面板显示剩余使用量充足的情况下： <img alt="Image" width="822" height="386.765625" src="https://private-user-images.githubusercontent.com/318925/558794128-4ef86ff5-08e3-4049-906b-28793602e8d3.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NzI4Njc3NDAsIm5iZiI6MTc3Mjg2NzQ0MCwicGF0aCI6Ii8zMTg5MjUvNTU4Nzk0MTI4LTRlZjg2ZmY1LTA4ZTMtNDA0OS05MDZiLTI4NzkzNjAyZThkMy5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwMzA3JTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDMwN1QwNzEwNDBaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT0wMjNkNmE2NDlhZWQ4MzEwZTFhM2U4MjVjNzliY2ZmZmZkODUxYjgyY2FlYWVkY2RiNjQwMGI0MDYzNGQwNWJhJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.ZqqMYVqDkJHvo5-DhQBcxjO1EXFRcjIm_-jA5Oq21eg"> Does this mean that there was likely a _different bucket_ of additional credits that are _not represented_ in the usage UI which the account was feeding off of? And that the behavior change (api limit reached, 1m models missing) indicates that this _other_ bucket has been disabled or depleted?这是否意味着账户可能依赖着另一组额外的配额，这些配额并未在使用情况界面中显示？而行为变化（API 限制达到、100 万模型缺失）表明这组_其他_配额已被禁用或耗尽？ This might not be debug-able if you are not also the admin of your account.若您并非账户管理员，此问题可能无法调试。

I think I found the reason, it may because the Extra usage is run out, you should recharge it and can be continue

<img width="1086" height="746" alt="Image" src="https://github.com/user-attachments/assets/8db09153-c1d7-4bed-a176-f3aad0577ffd" />

Ya, I had opus with 1m context selected by accident. This error message needs to be more clear.

<img width="494" height="372" alt="Image" src="https://github.com/user-attachments/assets/2a3e3b19-0902-4b64-a910-9c3548f1e024" />

Corroborating report + UX feedback (Claude Code v2.1.72, Max plan, macOS)

Independently hit the same issue and can corroborate the OP's follow-up analysis on the root cause (1M context billing above 200k tokens as extra usage).

My experience

• Opus 4.6 (1M) — rate limited
• Sonnet 4.6 (1M) — also rate limited (corroborates OP's update)
• Both models without 1M context — work fine
• Confirmed across two machines, three terminals (iTerm2, Ghostty, Terminal.app), and a fresh claude logout && claude login cycle
• Usage dashboard showed ample quota (session: 12%, weekly: 41%, Sonnet: 0%) — nothing indicated what was wrong

What I found in the local config

While troubleshooting, I noticed the following values in ~/.claude.json (unclear if related):

"hasAvailableSubscription": false,
"cachedExtraUsageDisabledReason": "out_of_credits"

Based on the OP's follow-up analysis, these values were likely accurate — my extra usage balance had hit $0, and 1M context requires extra usage credits above 200k tokens. However, these values persisted even after switching to non-1M models that should work fine under the regular subscription. It's unclear whether they would have eventually self-corrected or if they could cause issues in other contexts.

Screenshots

<img width="1950" height="870" alt="Image" src="https://github.com/user-attachments/assets/696b4cda-2e19-42a7-aa6b-f41a75d30719" />

UX improvement requests

The core issue here is that the system gives zero indication of what's happening. Some suggestions:

1. The error message needs to say what's actually wrong — "Rate limit reached" should be something like "1M context requires extra usage credits. Your balance is $0. Switch to standard context or add credits."
2. The usage dashboard should surface 1M context / extra usage status — there's no visible meter that explains why you're blocked when every other meter shows green.
3. Claude Code should auto-fallback to the non-1M variant when 1M is unavailable, or at minimum prompt the user to switch — instead of failing repeatedly with a generic error.
4. Proactive warning when approaching extra usage limits — e.g., "You've used 80% of your extra usage balance for 1M context." This would also help users who temporarily switch to 1M for a task and forget to switch back.

Related: #29882, #29579, #27179, #27535, #26496

I’ve started having this issue tonight after the 1M context became the default for Opus 4.6. Claude Max $200 plan.

got this issue in cli, but works in claude code desktop app

Bonjour, de ce que j'ai compris et réussis, il faut forcément activer l'extra usage et mettre de l'argent sur la version claude web pour avoir les models 1M

I noticed it seems to happen with my work/corporate account but not my personal Max 20x personal account. Both offer Opus 4.6 1M by default, but when I reach about 180k of context on some conversation with my work account, I get "API Error: The model has reached its context window limit.".

It seems hit or miss. Sometimes, I retry and it'll keep saying it reached it max context, and at other times, it goes through:

❯ /model ⎿ Set model to Opus 4.6 (1M context) (default) ❯ Yes please, and pay tech debt as you go to avoid carrying it over. ⎿  API Error: The model has reached its context window limit. ❯ . ● Let me first clean up tech debt from Slice 1, then proceed to Slice 2.

It makes me believe the Anthropic routing is sometimes being routed to Opus 4.6 200k context instead of the 1M context, even when 1M is set. Could this be based on server workload since it's a bit inconsistent? 🤔