Unclear how 1M context billing works on Max subscription
Status Closed — not planned
Maintainer reply None cached
Activity 15 comments · opened Feb 25, 2026 · closed Jun 2, 2026
Question
When selecting a 1M context model (e.g. claude-opus-4-6[1m]), Claude Code shows "Billed as extra usage." However, the actual billing behavior on a Max subscription is unclear.
Observed behavior (Max 5x plan)
- Activated
claude-opus-4-6[1m]in a session - Extra Usage ($24.35/$42.50) did not change over ~50 minutes of active use
- Session quota went from 8% → 67% in that time (much faster than normal)
- Weekly quota went from 26% → 31%
- Context was ~32% filled (~320K tokens)
Questions
- Does 1M context on Max consume Extra Usage or regular quota? The "Billed as extra usage" message suggests the former, but our Extra Usage balance didn't change. Regular quota consumed faster instead.
- Why is quota consumed faster? Two possible mechanisms — unclear which (or both) apply:
- Larger context = more tokens per message. With ~320K context, every roundtrip naturally sends more tokens. This alone would explain faster quota drain without any special multiplier.
- 2x rate multiplier. API pricing doubles when a request exceeds 200K input tokens. Does Max subscription apply a similar penalty multiplier on top of the naturally higher token count?
- What does "Billed as extra usage" actually mean for Max subscribers? If it just means faster quota consumption (not actual Extra Usage charges), the message is misleading.
- Is there documentation on this? We couldn't find clear answers for how the >200K API pricing threshold translates to subscription-based usage.
Environment
- Claude Code (latest)
- Max 5x subscription
- Model:
claude-opus-4-6[1m]
🤖 Generated with Claude Code
Showing cached comments. Read the full discussion on GitHub ↗
14 Comments
Found 3 possible duplicate issues:
This issue will be automatically closed as a duplicate in 3 days.
🤖 Generated with Claude Code
Not a duplicate. This issue is specifically about the 1M context window as specified by Model: claude-opus-4-6[1m]
To clarify the 1M context billing based on Anthropic's pricing docs (as of March 2026):
How it works:
| Model | Standard Input | Extended Input (>200K) | Standard Output | Extended Output (>200K) |
|-------|:---:|:---:|:---:|:---:|
| Opus 4.6 | $5/MTok | $10/MTok (2x) | $25/MTok | $37.50/MTok (1.5x) |
| Sonnet 4.6 | $3/MTok | $6/MTok (2x) | $15/MTok | $22.50/MTok (1.5x) |
Key gotcha: When you cross 200K input tokens, ALL tokens get billed at the premium rate — not just the tokens above 200K. So going from 199K to 201K tokens doesn't just make the extra 1K more expensive; it makes all 201K tokens 2x on input.
On Max subscription: The 1M context models consume your session/weekly quota at a higher rate (which matches what you observed — 8% to 67% session quota). The "extra usage" meter tracks API-billed overages, which is separate from the quota system.
Practical advice: Only use
[1m]models when you genuinely need >200K context. For most coding sessions, the standard 200K context is sufficient. If you're hitting context limits, use/compactto compress history rather than upgrading to 1M.I documented the full pricing breakdown including 1M context, Fast Mode, Bedrock/Vertex pricing, and cost optimization strategies at claude-cost-optimizer/guides/06-access-methods-pricing.md.
Thanks for your answer. That's the official numbers for API usage. It would be nice to know for sure if the same multipliers (2x / 1.5x) also apply to subscriptions.
I disagree about [1m] usage. It is now my standard setting. Otherwise, when hitting the 200k limit the conversation is autocompacted. I have found the results to be unreliable after autocompact. And you are only billed extra when you go over 200k.
So which mean under the max plan , we are free to use 1m context but the cost are using our weekly session quota ? So meaning we are pay for the 1m context via using the weekly session quota ? Is my understanding correct ? Thanks
If you disable extra usage in https://claude.ai/settings/usage then claude code will say that opus 1m isn't available for your account. That makes me think that it behaves exactly as it says - 1m context is billed as extra usage, meaning it's not included in our plans. Happy to be proven wrong though
[](https://github.com/anthropics/claude-code/blob/main/CHANGELOG.md)
In the new changelog , it showing [Added 1M context window for Opus 4.6 by default for Max, Team, and Enterprise plans (previously required extra usage)]
https://claude.com/blog/1m-context-ga
so i think is free now
Adding another data point: I'm on the Max 200 plan running Claude Code v2.1.76, and the
/modelpicker still shows Opus 4.6 (1M context) as "Billed as extra usage."This is despite the v2.1.75 release notes stating:
At minimum, the display label in the model picker appears to be incorrect for Max subscribers.
i had try it and is free now , cannot say is free , is under our usage calculation now
<img width="1034" height="199" alt="Image" src="https://github.com/user-attachments/assets/5691e279-9111-449b-8890-71e908956f8b" />
This is a common source of confusion. Here's the key detail:
When your input exceeds 200K tokens, all tokens in the request get billed at the 1M context rate -- not just the tokens above 200K. So crossing from 199K to 201K input tokens doubles your entire input cost.
Example with Opus 4.6:
Strategies to stay under the threshold:
Full explanation with worked examples: https://github.com/Sagargupta16/claude-cost-optimizer/blob/main/guides/01-understanding-costs.md#long-context-pricing-1m
I've done extensive analysis on this exact problem. Using ccusage_go (open-source Claude Code usage tracker), I found that Cache Read tokens consumed 97.7% of my session costs — API actual cost was $1.47, total billed cost was $64.98 (a 44x markup). Cache also degrades instruction following in long sessions, which I documented with per-turn JSONL analysis.
Full write-up with data, community issue references, and Claude Code's own self-analysis report:
https://blog.sd.idv.tw/en/posts/2026-03-25_claude-code-cache-trap/
Tool: https://github.com/SDpower/ccusage_go
Yeah, I think the confusing part is that with the 1M context models the “size” of a turn isn’t really obvious anymore.
Even if the visible prompt is small, it can still be resending a lot of prior context or tool output under the hood, so the quota % drops way faster than expected.
In your case, is it happening right from the first few turns, or does it start ramping once the session builds up?
the 'Billed as extra usage' label is misleading on Max. mechanics: 1M context sends more input tokens per request (your 320K context = roughly 10x normal), AND Anthropic applies a 2x rate multiplier on >200K input requests (matches the API pricing tier). on Max both drain regular session quota faster, neither moves the Extra Usage balance. that's how session went 8% to 67% in 50min while Extra Usage stayed pinned at $24.35. the label refers to the API pricing tier, not your subscription bucket. UX bug, not a billing bug.
Closing for now — inactive for too long. Please open a new issue if this is still relevant.