Bring back "opusplan", why did you remove it in v2.0.0?

[PaulRBerg]

Preflight Checklist

• [x] I have searched existing requests and this feature hasn't been requested yet
• [x] This is a single feature request (not multiple features)

Problem Statement

opusplan was an option for /model that was available in v1.0.x, which enabled us to use Opus for planning and Sonnet for implementation. The CLI was automatically switching between Opus and Sonnet without us having to manually toggle between them after the planning session is done.

Why has it been removed in v2.0.0?

Proposed Solution

Bring opusplan back!

Priority

High - Significant impact on productivity

Feature Category

Configuration and settings

[nfarina]

I rely heavily on Opus Plan mode myself. I'm curious if this change was made because Anthropic believes Sonnet 4.5 is a better planner than Opus 4.1?

[PaulRBerg]

if I were to guess, it wasn't an intentional change. Prolly just a vibe-coded removal

this is a huge downgrade in terms of UX, guys, PLEASE bring opusplan back!

[wpfleger96]

FWIW you can still still set this in your ~/.claude/settings.json:

"model": "opusplan"

And it still works normally with Opus for plan and Sonnet for act... I'm guessing it was erroneously removed as a UI option

[mattslight]

I'm all up for innovation at the speed of thought, but if it's not tested... Please please please Claude team, do not rush. You are already better than OpenAI Codex, don't make me switch because you drop the ball and rush releases.

Thank you for the fix @wpfleger96

[WtheFourth]

Even when using /model opusplan or setting "model": "opusplan" in my settings.json, when I ask it what model it's using it always appears to at least think it's using Opus. The model on my statusline is what I expected it to use (Sonnet 4.5).

[PaulRBerg]

I set opusplan in settings.json, and this is what I see when I run claude:

• Opus 4.1 at the top
• Sonnet 4.5 in the status line

Very confusing and I'm not sure now if @wpfleger96's fix actually works.

[WtheFourth]

@PaulRBerg I'm reasonably sure it doesn't; I asked mine a few times when swapping through the various modes what model it was using and it stuck with Opus every time regardless of whether I was using Plan or not.

[PaulRBerg]

then this is a bug and it should be fixed ASAP

I will go back to v1.0.128 in the meantime

[christopher-buss]

According to Anthropic's own benchmarks, Sonnet 4.5 is currently a better model than Opus 4.1 across the board. This might be a case that it's removed as an option until a better version of Opus is released.

[wpfleger96]

It seems like the UI is more buggy than I initially noticed... I may have an idea what's going on here by inspecting the debug logs from my custom statusline script

First, I think setting opusplan in ~/.claude/settings.json is "working" in that it uses Opus for plan mode and Sonnet for act mode. However, which Opus and Sonnet models exactly are being used is a different story... I can confirm that setting opusplan results in my plan mode using Opus 4.1 and act mode using Sonnet 4. But I believe that's because I have this config in my settings.json too:

  "env": {
    "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-20250514",
    "ANTHROPIC_DEFAULT_OPUS_MODEL": "claude-opus-4-1-20250805"
  }

So at least based on the payloads that Claude Code is sending to my status line script, it seems like it's working as I intended i.e. plan mode uses Opus 4.1 and act mode uses Sonnet 4. But hypothetically if opusplan mode is set, but ANTHROPIC_DEFAULT_SONNET_MODEL/ANTHROPIC_DEFAULT_OPUS_MODEL are NOT, it's up to Anthropic which version exactly of those models will be used in the session... so I'm wondering if that's leading to some of the discrepancies here.

Also FWIW, most LLMs are notoriously bad at knowing what model they are... so using the /status slash command, or checking the payloads sent to status line scripts, will be much more accurate in denoting the current model being used. I'm not sure of this, but I'm assuming that the outputs of /status and the status line payloads will be more accurate than the UI values given the issues already identified in the 2.0.0 UI

[wpfleger96]

I just confirmed that if I set opusplan in settings.json, but remove "ANTHROPIC_DEFAULT_SONNET_MODEL": "claude-sonnet-4-20250514", then plan mode still uses Opus 4.1 but act mode now uses Sonnet 4.5 instead of Sonnet 4

It seems like this initial top line only gets rendered when the session starts and doesn't update in real time:

Also for @PaulRBerg 's issue, there's a rough edge with opusplan mode that it doesn't automatically start your session in plan mode... which you can see by bypass permissions on being shown in your status line. If you were in plan mode you'd see plan mode on in that same place instead. If you use shift+tab to cycle to plan mode, you should see the model correctly update to Opus in the status line

At least from my experience, opusplan only works (i.e. uses Opus to plan, presents the plan to the user where they can approve or deny, and if they accept the plan it auto switches to Sonnet for implementation) if the session is already in "plan" mode set via shift+tab. It's definitely confusing 🤷

[dicksontsai]

Hi all, thanks for your feedback! As you might imagine, we are open to changing the UI based on the feedback we hear over the next few days, but I want to add some clarifications in the meantime.

1. We removed it from the model selector UI intentionally because we have found Sonnet 4.5 to be generally better than Opus 4.1, especially if you enable thinking for Sonnet 4.5. Why not give it a shot? Use Tab to control thinking.
2. The opusplan setting still works. You can manually set it in ~/.claude/settings.json as pointed out above or use /model opusplan.
   1. As pointed out above, ANTHROPIC_DEFAULT_OPUS_MODEL and ANTHROPIC_DEFAULT_SONNET_MODEL control which "opus" and "sonnet" models opusplan uses. In v2.0.0, the defaults for "opus" and "sonnet" are Opus 4.1 and Sonnet 4.5 respectively.
   2. There is a UI bug where opusplan does not show up well after our change (shows up as a "custom model"). We will fix this issue in v2.0.1.
3. Do not ask the model what model it is. You can use Ctrl+o to see which model fulfilled your request. It's a question best handled by the application (deterministic), not by a language model (different context -> different answers).
4. The logo is rendered statically. It won't update dynamically based on the current model setting. For the latest status, customize a status line: https://docs.claude.com/en/docs/claude-code/statusline
   1. For Opus Plan Mode, the model used is Sonnet unless you are in plan mode (Shift+Tab on Mac/Linux/some Windows, Alt+M on some Windows with older Node). Status Line should reflect the model that is used.

[wowitsjack]

Hi all, thanks for your feedback! As you might imagine, we are open to changing the UI based on the feedback we hear over the next few days, but I want to add some clarifications in the meantime.

1. We removed it from the model selector UI _intentionally_ because we have found Sonnet 4.5 to be generally better than Opus 4.1, _especially_ if you enable thinking for Sonnet 4.5. Why not give it a shot? Use `Tab` to control thinking.

Is this why you removed 'think' 'think hard' 'think very hard'? :/

Discord is already lit up with folks trying to revert to get this critical functionality back.

[mattslight]

Thinking seems to help. But simple UI tasks like moving the position of two CTAs it struggled with simple comprehension tasks. What would have taken one attempt took half a dozen and honestly the frustration factor was real. Previously I was doing complex refactors with Claude. Right now that seems like light years away.

Unfortunately I don't know what my model settings were on the 1.x branch so it's hard to do a scientific test.

Right now it feels like Claude's gone from 5 year experienced dev to kindergarten since 2.0 release.

The new features and the crab logo are cool and all but if the core feature of writing code suffers then it will push me and I assume others to use competitors tooling. $100 a month per dev for Max so it's not a small cost. More credits on Opus might help?

On Tue, 30 Sep 2025 at 07:53, Jack @.***> wrote:

wowitsjack left a comment (anthropics/claude-code#8358) https://github.com/anthropics/claude-code/issues/8358#issuecomment-3349842848

Hi all, thanks for your feedback! As you might imagine, we are open to changing the UI based on the feedback we hear over the next few days, but I want to add some clarifications in the meantime.

1. We removed it from the model selector UI intentionally because we have found Sonnet 4.5 to be generally better than Opus 4.1, especially if you enable thinking for Sonnet 4.5. Why not give it a shot? Use Tab to control thinking.

Is this why you removed 'think' 'think hard' 'think very hard'? :/

Discord is already lit up with folks trying to revert to get this critical functionality back.

— Reply to this email directly, view it on GitHub https://github.com/anthropics/claude-code/issues/8358#issuecomment-3349842848, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIGDZ5MRURLOP3C5IJRDZD3VH5EDAVCNFSM6AAAAACH2HENDWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNBZHA2DEOBUHA . You are receiving this because you commented.Message ID: @.***>

[FredHasselot]

Could you please take a look at my related issue #8449?

I believe the automatic model selection bug is directly related to this opusplan removal. The bug caused my system to use expensive models (Opus) for trivial tasks without my control, leading to excessive token consumption.

I found the workaround (manually setting opusplan in settings.json), which seem to resolved the issue. However, I'm now unfairly rate-limited due to a system bug.

My questions:

1. Is this overconsumption issue related to the opusplan removal and UI inconsistencies?
2. Have you received similar reports of excessive token consumption due to this bug?
3. What does the team plan to do to reset rate limits for users affected by this bug?

I understand bugs happen, but paying customers shouldn't be penalized with rate limits caused by system issues beyond their control. I'd appreciate the team's position on compensating affected users.

[FredHasselot]

After use, it is possible that the diagnosis is rather than that it happens during the compact of the conversation

[PaulRBerg]

Thanks for the reply @dicksontsai

I played with Sonnet 4.5 for a while today, and yeah, it does seem a little better than Opus 4.1 in plan mode - or at least as good as Opus 4.1

May I suggest to change the wording in the model selection UI? It's currently worded like this:

---

Which makes it seem like Sonnet 4.5 is like Sonnet 4.0, i.e., an execution guy. But Sonnet 4.5 is both a judicious planner and an execution guy. The current wording doesn't clearly communicate this.

[maksymilian-majer]

You can use old models with the --model CLI option, but it seems that it will use Sonner 4.1 instead of 4.5 out of planning.

[mattslight]

Claude 2.0.2 seems to have restored the option for OpusPlan.

On Thu, 2 Oct 2025 at 17:13, Maksymilian Majer @.***> wrote:

maksymilian-majer left a comment (anthropics/claude-code#8358) https://github.com/anthropics/claude-code/issues/8358#issuecomment-3361174755

You can use old models with the --model CLI option, but it seems that it will use Sonner 4.1 instead of 4.5 out of planning.

— Reply to this email directly, view it on GitHub https://github.com/anthropics/claude-code/issues/8358#issuecomment-3361174755, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIGDZ35WW6I6MXGMQRLOC33VUQJBAVCNFSM6AAAAACH2HENDWVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZTGNRRGE3TINZVGU . You are receiving this because you commented.Message ID: @.***>

[freegurl]

You can use old models with the --model CLI option, but it seems that it will use Sonner 4.1 instead of 4.5 out of planning.

I've updated my VS Code extension but it still has the old references (I was also using the CLI too with --model). It was working last week but this today when using Opus Plan Mode it was defaulting my session to Opus and maxed my Opus limit for the week. (Max Plan)

[priyashpatil]

It seems like bug. As a max plan user I can use any model mentioned in docs by setting model manually and it works. It just model selection is broken.

[lpalbou]

According to Anthropic's own benchmarks, Sonnet 4.5 is currently a better model than Opus 4.1 across the board. This might be a case that it's removed as an option until a better version of Opus is released.

except in real life practice, it's absolutely not the case. it might have been fine tuned on benchmarks, but Sonnet 4.5 is one of the worst model created so far. It is however fast and probably cheap for them to run. Reasoning is extremely limited. The whole claude code experience has also been tuned down (eg no more way to trigger thinking mode with keywords : think, think carefully, ultrathink etc).

Definitely a downgrade. "opusplan" was amazing in comparison.

[mintmcqueen]

@dicksontsai : suggestion for the future. Create /model: planmode option that allows user to define model for planning and model for execution. More choice = more happiness / fewer complaints.

[john-savepoint]

Are we getting this feature or not? This is over a month old, and it's asking for functionality that already existed in a previous version. So did something break fundamentally?

[PaulRBerg]

did something break fundamentally?

Anthropic's runway

(As a result of running Opus in the cloud)

[lzilioli-prism]

Yeah this mode should be restored–I miss it dearly. Played with the workarounds in the thread, but it doesn't seem to be very reliable. Hard to tell if its working at all.

[cruftyoldsysadmin]

@claude You have more money than many countries. Do better.

[PaulRBerg]

Hey @dicksontsai & Anthropic team, now that Opus 4.5 is live and also available Claude Code v2.0.51, can you please, please bring opusplan back? 🙏

I don't want to have to manually switch between models using /model.

The Opus-for-planning and Sonnet-for-execution combo was a killer feature.

[johnnyApplePRNG]

Seriously... come on guys... give us opusplan!!! What is the holdup?!

[Drew-Goddyn]

The revenue earned from picking /model opus and forgetting to switch to sonnet on execution far exceeds the tears in this thread.