Model Alias Mapping

TULIP currently deploys models optimized for three different tasks. Models are identified by aliases for convenience.

Capability	Alias	Current Backing Model	Notes
Chat	`chat`	`google/gemma-4-31B-it-qat-w4a16-ct`	Default chat model; tool-calling format may vary by serving/parser configuration
Coding	`code`	`Qwen/Qwen3.6-27B-FP8`	Supports code generation and reasoning; verified in reference tests for structured OpenAI-style tool calling (`message.tool_calls`)
Embedding	`embed`	`Qwen/Qwen3-Embedding-8B`	4096-dimensional embeddings for retrieval and semantic search

Query Runtime Models (Optional)

This check is still useful for debugging and environment validation. However, the serving layer may return alias-oriented output (and not real model names).

#!/usr/bin/env bash

SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
if [ -f "$SCRIPT_DIR/../.env" ]; then
  source "$SCRIPT_DIR/../.env"
fi

: "${TUDELFT_TULIP_API_KEY:?Set TUDELFT_TULIP_API_KEY in .env or environment}"

for route in chat code embed; do
  curl -s --fail --show-error "https://api.tulip.tudelft.nl/${route}/v1/models" \
    -H "Authorization: Bearer ${TUDELFT_TULIP_API_KEY}" \
    -H "Content-Type: application/json" | jq -r '.data[].id'
done

Run it with:

bash examples/curl-check-models.sh

Important Notes

Update channel: Alias-to-model mapping changes are communicated via the TULIP mailing list.
Prefer aliases: Use chat, code, and embed in your clients — don't hardcode backing model IDs, as they may change.
Tool-calling behavior: For current limitations and expected output format, see the Tool Calling tutorial.