The recent VS Code extension update has noticeably degraded the experience.
The agent now becomes unresponsive at times and needs a reload, which really breaks flow. More frustratingly, the context limit seems to fill up much faster on the same project that was working fine just days ago. Nothing major changed on my side, so this feels like a backend or token allocation shift.
There's a real hobbyist vs professional distinction with Claude Code. For professionals, including when I use it at work, we're generally super happy to have Claude spawn as many subagents as possible and burn more tokens to get a better result. Hobbyist users on a $20/month plan, though, generally want more conservative behavior.
It's hard for Anthropic to cater to both sets of users with one model.
I don't think that's what this issue is talking about. I have the Max $200/mo plan and have noticed starting yesterday that my quota drains much much faster, to the point I'm about to use the $50 credit Anthropic gave away to everyone.
True enough. But to be clear, that's a separate issue from what users are reporting here.
Both hobbyists and professionals are understandably frustrated that tokens are being consumed quickly without justification, or at least in ways that seem entirely avoidable.
I have MAX and have been using Opus 4.6 heavily for my day job which is 100% agentic programming, and my usage numbers have not changed meaningfully since Opus 4.6 came out
Same here. Both $20 and $100 finished fast. Never hit a limit after dishing out the $200. Explore() sometimes prints 90k token usage which scares me, but so far it is consequence free.
At 8 hours a day 5 days a week I never hit the limit on my $100 MAX plan. People must be running crazy autonomous workflows or something because I'm nowhere near hitting my limit, ever
Each time I've dug into this for someone, it's because they're filling up their context window with a bunch of tokens before any real work even starts.
Highly encourage people having issues to do /context and start removing heavy things. It's usually some sprawling MCPs they rarely use, or huge CLAUDE.md files they generated or cargo-culted from someone else.
I'm not suggesting these are the only ways to hit the limits, it's just (so far) almost always the answer when someone hits the limits doing something that I wouldn't expect to be problematic.
My quality of usage with Claude has degraded heavily since last week of December that I've stopped using Claude entirely now and mostly found a Codex suitable, and goes much further. I was maxing out usage on Claude Max 5x in absurd ways once I started using MCP features heavily, and even when not found myself constantly hitting limits through January.
The final nail was them offering a $50 credit toward overage use that within a half hour of enabling maxed out and began digging into. It's become almost predatory now, and I have no way to quantify the actual usage I'm getting from it other than it burns now at an alarming rate.
Since I've stopped using Claude, I ultimately landed on Codex where for my usage, where I'm easily getting 4x less quota usage from it than Claude for the same period of heavy use. I keep it as a backup now if Codex gets stuck on something, but I'm annoyed enough to stop paying all together.
Update Claude, turn on all of the MCPs you've been using, start a new Claude session from scratch in an empty folder.
Run /context.
Observe that your MCPs are killing a sizable chunk of the usable context window.
The utility here is that it'll break down exactly which MCPs are consuming how many tokens, just in tool descriptions. Then you can decide if that's worth it to you, even if you continue using Codex or OpenCode or etc.
selridge | 18 hours ago
Vaslo | 18 hours ago
selridge | 18 hours ago
Yeah.
What's the 'news'?
stingraycharles | 18 hours ago
I’ll believe it when I see actual facts, e.g. actual token counts (which is relatively easy to capture if you use mitmproxy or something like that).
For all I know this guy has a 5000 line CLAUDE.md
SoftTalker | 18 hours ago
hinkley | 17 hours ago
I should update my notes.
[OP] behnamoh | 17 hours ago
selridge | 7 hours ago
That’s the implied question. It’s an issue on a piece of software. There’s more than 10,000 of them.
Why is it news?
sparin9 | 18 hours ago
The agent now becomes unresponsive at times and needs a reload, which really breaks flow. More frustratingly, the context limit seems to fill up much faster on the same project that was working fine just days ago. Nothing major changed on my side, so this feels like a backend or token allocation shift.
visarga | 16 hours ago
thurn | 17 hours ago
It's hard for Anthropic to cater to both sets of users with one model.
[OP] behnamoh | 17 hours ago
xwowsersx | 17 hours ago
Both hobbyists and professionals are understandably frustrated that tokens are being consumed quickly without justification, or at least in ways that seem entirely avoidable.
pdntspa | 16 hours ago
nurettin | 16 hours ago
pdntspa | 5 hours ago
mh- | 3 hours ago
Highly encourage people having issues to do /context and start removing heavy things. It's usually some sprawling MCPs they rarely use, or huge CLAUDE.md files they generated or cargo-culted from someone else.
I'm not suggesting these are the only ways to hit the limits, it's just (so far) almost always the answer when someone hits the limits doing something that I wouldn't expect to be problematic.
nickvec | 16 hours ago
rognjen | 16 hours ago
bastard_op | 15 hours ago
The final nail was them offering a $50 credit toward overage use that within a half hour of enabling maxed out and began digging into. It's become almost predatory now, and I have no way to quantify the actual usage I'm getting from it other than it burns now at an alarming rate.
Since I've stopped using Claude, I ultimately landed on Codex where for my usage, where I'm easily getting 4x less quota usage from it than Claude for the same period of heavy use. I keep it as a backup now if Codex gets stuck on something, but I'm annoyed enough to stop paying all together.
mh- | 4 hours ago
Run /context.
Observe that your MCPs are killing a sizable chunk of the usable context window.
The utility here is that it'll break down exactly which MCPs are consuming how many tokens, just in tool descriptions. Then you can decide if that's worth it to you, even if you continue using Codex or OpenCode or etc.