How to stop openrouter/deepseek/deepseek-r1 from thinking in thousands of tokens ?

[GeorgeFlorian]

Issue

How to stop openrouter/deepseek/deepseek-r1 from thinking in thousands of tokens ?
I input a prompt, it thinks in 100+ lines with code and all that shit and then it ask to edit files and then it again uses token to write the same code from the thinking process.

~/.aider.model.setting.yml:

name: openrouter/deepseek/deepseek-r1
  extra_params:
    top_p: 0.95
  use_temperature: 0.6

~/.aider.conf.yml

model: openrouter/deepseek/deepseek-r1
architect: true
editor-model: openrouter/anthropic/claude-3.5-sonnet
auto-accept-architect: false
auto-commits: false

cache-prompts: false
stream: true

dark-mode: true

lint: true

## Set the thinking token budget for models that support it (default: not set)
# thinking-tokens: 100

I have tried thinking-tokens: 100, it does not work.
I have tried remove_reasoning: think which is a setting that aider can't start with. (#3008 (comment))

Is there no fix to this issue ?
Should I just change from R1 ?

Version and model info

Aider v0.81.1
Main model: openrouter/deepseek/deepseek-r1 with architect edit format, prompt cache, infinite output
Editor model: openrouter/anthropic/claude-3.5-sonnet with editor-diff edit format
Weak model: openrouter/deepseek/deepseek-chat
Git repo: .git with 75 files
Repo-map: using 4096 tokens, auto refresh

[LangLangBart]

How to stop openrouter/deepseek/deepseek-r1 from thinking in thousands of tokens ?

If you mean by 'stop' that the reasoning should not be included in the response, then the following should work:

- name: openrouter/deepseek/deepseek-r1
  extra_params:
    top_p: 0.95
    reasoning:
      # The model will still use reasoning, but it won’t be shown in the response.
      exclude: true
  use_temperature: 0.6

• Ref: https://openrouter.ai/docs/use-cases/reasoning-tokens

If you mean with 'stop' to not reason at all, then the reasoning.effort or reasoning.max_tokens did not help.

---

The reasoning field with openrouter appears to be quite new (around late February) according to the web archive, which might explain why it's not mentioned in the aider model settings documentation.

17/Feb/25 - include_reasoning

• https://web.archive.org/web/20250217150148/https://openrouter.ai/docs/use-cases/reasoning-tokens

26/Feb/25 - reasoning.exclude

• https://web.archive.org/web/20250226060225/https://openrouter.ai/docs/use-cases/reasoning-tokens

[GeorgeFlorian]

@LangLangBart

- name: openrouter/deepseek/deepseek-r1
  extra_params:
    top_p: 0.95
    reasoning:
      # The model will still use reasoning, but it won’t be shown in the response.
      exclude: true
  use_temperature: 0.6

I have added exactly this configuration in ~/.aider.conf.yml and there has been no change, it still shows me the Thinking process when responding.

[LangLangBart]

I have added exactly this configuration in ~/.aider.conf.yml and there has been no chang

The config with reasoning.exclude should go into the ~/.aider.model.setting.yml.

❯ cat ~/.aider.model.settings.yml
---
- name: openrouter/deepseek/deepseek-r1
  extra_params:
    top_p: 0.95
    reasoning:
      # The model will still use reasoning, but it won’t be shown in the response.
      exclude: true
  use_temperature: 0.6

❯ cat ~/.aider.conf.yml
---

model-settings-file: .aider.model.settings.yml
alias:
  - r1:openrouter/deepseek/deepseek-r1

env-file: /Users/paria/.dotfiles/aider/.env

Run a test command with --verbose and make sure the reasoning.exclude is shown

aider --no-git --model r1 --message 'just say hi' --verbose --edit-format ask

...
ASSISTANT Ok.
-------
USER just say hi
kwargs:
{
    "model": "openrouter/deepseek/deepseek-r1",
    "stream": true,
    "temperature": 0.6,
    "top_p": 0.95,
    "reasoning": {
        "exclude": true
    },
    "timeout": 600
}
Hi! 👋 How can I assist you with code analysis today?                                                                                                                                         


Tokens: 81 sent, 14 received.