v4.1 flash's input doesn't have ["image"]: Pi

Quick fix

Upgrade to v4.0.0 and add DeepSeek again from the preset, or manually set the model input to ["text", "image"].

Symptom

Error outputCopy

v4.1 flash's input doesn't have ["image"]

Cause

The Pi DeepSeek preset in v3.20.4 uses the old `deepseek-v4-flash` entry, which supports text input only. This was fixed in v4.0.0: the preset now uses `deepseek-flash` (V4.1 Flash) and declares `"input": ["text", "image"]`.

DeepSeek providers added in v3.20.4 are not updated automatically after upgrading because presets are applied only when a provider is added.

Fix

1. 1

   Upgrade to v4.0.0.

2. 2

   If the DeepSeek provider was added in v3.20.4, add DeepSeek again from the preset after upgrading.

3. 3

   Or manually set the model's `input` to `["text", "image"]`.

Affected Versions

ToolPi

Versionv3.20.4（受影响；v4.0.0 已修复）

PlatformsLinux

Source Issues

This page is distilled from 1 real issues

• #7859Pi + deepseek v4.1 flash doesn't have "input": ["image"] in the defalut setting

FAQ

(No FAQ yet)

Related problems

<thinking> tags output as body text in Claude CodeChanging the provider's apiFormat from openai_chat to anthropic, if supported upstream, can bypass inline <thinking> leakage.1 issues mergedHowever, this session isn't eligibleUpdate to a version with the #7893 fix; it writes `CLAUDE_CODE_AUTO_MODE_SERVER` to settings.json when switching proxies.1 issues mergedsession isn't eligibleFor a non-Anthropic upstream or format conversion, set CLAUDE_CODE_AUTO_MODE_SERVER=0 in the provider env.2 issues merged