[BUG] 1M context unavailable on Max plan — Windows 10

Found 3 possible duplicate issues:

1. https://github.com/anthropics/claude-code/issues/34333
2. https://github.com/anthropics/claude-code/issues/34585
3. https://github.com/anthropics/claude-code/issues/36773

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

None of the solutions in other bug reports worked for me, not the log in, log out, not using the ENV, nothing.

I don't have the option to select 1M context at all switching between different models to figure it out. Nothing that I do in any way shows an option to get that one million context back.

I can confirm that till 8 Hours Ago, I had 1M Context Available. Now When I resumed work, 1m Context is gone, my previous conversation started throwing Prompt is too long, When I check available models 1M Opus 4.6 is no longer visible. This is definitely a regression.

I can confirm that till 8 Hours Ago, I had 1M Context Available. Now When I resumed work, 1m Context is gone, my previous conversation started throwing Prompt is too long, When I check available models 1M Opus 4.6 is no longer visible. This is definitely a regression.

That's essentially what happened to me. I went to bed two days ago and woke up to the smaller context. I do however have 1M in Cowork but it's very glitchy.

My Claude says your issue is of the same type as issues #38238 and #39000. You should cross-reference them. Besides seeing if you can find a workaround together, it also helps get Anthropic's attention on your issue.

The issue isn't a bug: it's Anthropic stratifying their Sonnet[1m] model behind a paywall, even when you've paid for a Max 20x plan. What's worse is we as users never expected this to creep beyond the Max 20x plan (which is technically the ceiling) and for API platform.anthropic.com economics to creep into a claude.ai product use.

The stop-gap fix is running 1m context on an older model: /model claude-sonnet-4-5-20250929[1m]

Anthropic and Claude.ai team have a bigger question to respond to:

Anthropic has contaminated Claude.ai with API economics and stratified the sonnet[1m] model for Max 20x users. What does Max 20x actually include now, and where is that documented?

Same issue here. CLI v2.1.100, Linux, Max plan.

/model shows only three options (Sonnet 4.6, Opus 4.6, Haiku 4.5) with no 1M variants. Previously had 1M context available.

Additional wrinkle: the model also silently switched from Opus to Sonnet without my action (previously reported in #13242 / #25675). Combined with the missing 1M option, this meant a session autocompacted at ~200K instead of continuing to 1M.

Two separate bugs compounding:

1. 1M context options missing from /model picker despite Max plan
2. Silent model downgrade from Opus to Sonnet without notification

Update: Resolved by deleting ~/.claude/.credentials.json and re-authenticating. The stale OAuth token was likely issued while travelling internationally with frequently changing IP addresses/locations. Fresh token from a stable IP restored Max plan entitlements — Default now correctly shows Opus 4.6 (1M context).

Root cause appears to be that entitlement state baked into the OAuth token doesn't update when network conditions change, and there's no mechanism to detect or notify the user of degraded entitlements.

+1, and the underlying cause @harrisonaedwards's fix resolves is that the stale token is literally missing the entitlement claim, not just out of date. hit this today on wsl2 w/ cc 2.1.114, stable ip, never travelled, max 20x. a plain claude login from wsl2 minted a token that looked like:

{
  "scopes": ["user:inference"],
  "subscriptionType": null
  // no rateLimitTier field
}

after rm ~/.claude/.credentials.json && claude login, the new token had subscriptionType: "max", rateLimitTier: "default_claude_max_20x", and /model claude-opus-4-7[1m] worked instantly.

quick diagnostic before burning a re-auth cycle @canucklehead2001:

python3 -c "import json; print(json.load(open('$HOME/.claude/.credentials.json')))"

if rateLimitTier is missing or subscriptionType is null, that's the bug. server denies 1m regardless of actual plan. not ip-related btw, the oauth exchange just sometimes drops the subscription claim on first issuance.

Closing for now — inactive for too long. Please open a new issue if this is still relevant.

This issue has been automatically locked since it was closed and has not had any activity for 7 days. If you're experiencing a similar issue, please file a new issue and reference this one if it's relevant.