Thinking Mode

View the reasoning process of DeepSeek, Gemma 4, and Qwen3 models with thinking blocks.

Thinking mode exposes the model's internal reasoning process as a separate response channel, so you can show a "thinking" bubble in your UI before the final answer.

Supported models#

Gemma 4 (E2B, E4B) — ModelType.gemma4
DeepSeek R1 — ModelType.deepSeek
Qwen3 0.6B — ModelType.qwen3; generates thinking by default, tags are stripped when isThinking: false.

Enable it with isThinking: true and the matching ModelType.

The reasoning channel is parsed per ModelType, and ModelType.general has no parser at all. Models that reason but run as general — SmolLM3 3B, Phi-4 Mini Reasoning — emit no ThinkingResponse, and their thinking tags are not stripped either: the raw blocks arrive inside the answer as ordinary TextResponse tokens. Strip them yourself, or don't advertise a thinking UI for those models.

Handling thinking responses#

The model emits a ThinkingResponse (with response.content) for its reasoning, alongside regular TextResponse tokens for the final answer:

chat.generateChatResponseAsync().listen((response) {
  if (response is ThinkingResponse) {
    // Model's reasoning process
    print('Thinking: ${response.content}');
    _showThinkingBubble(response.content);
  } else if (response is TextResponse) {
    // The final answer
    print('Text token: ${response.token}');
  }
});

You can also create a thinking message manually:

final thinkingMessage = Message.thinking(text: "Let me analyze this problem...");

Platform support#

Platform	Thinking Mode
Android	✅ Full with `.litertlm`; tag-based models only with `.task`
iOS	✅ Full with `.litertlm`; tag-based models only with `.task`
Desktop (macOS/Windows/Linux)	✅ Full
Web	⚠️ Qwen3 tag-based reasoning only

On Web, core can split Qwen3's emitted <think>...</think> tags into ThinkingResponse because that parser is platform-independent. Gemma 4 is a different path: the .litertlm engine passes extra_context and filter config, but the measured web_thinking_limitation_test.dart still receives only TextResponse, so its thinking channel is unsupported. MediaPipe Web has no thinking API, ONNX ignores enableThinking (on Web and native alike, though core still parses <think> tags a model emits), and the catalog's DeepSeek R1 .task model has no Web entry.

Advanced: ModelThinkingFilter#

For custom inference implementations, ModelThinkingFilter cleans model outputs — removing model-specific tokens. This is handled automatically by the chat API, but is available if you need it:

import 'package:flutter_edge_ai/flutter_edge_ai.dart';
import 'package:flutter_edge_ai/core/extensions.dart';

String cleanedResponse = ModelThinkingFilter.cleanResponse(
  rawResponse,
  isThinking: true,
  modelType: ModelType.deepSeek,
  fileType: ModelFileType.task,
);

// It removes the reasoning blocks (for these model types even when
// isThinking is false):
// - <think>...</think> (DeepSeek, Qwen, Qwen3)
// - <|channel>thought\n...<channel|> (Gemma 3 / Gemma 4 types)
// and trims whitespace. Turn markers (<end_of_turn>, <|im_end|>) are stripped
// only for .bin / .tflite files — on .task and .litertlm the runtime already
// ends the turn.