WOW! We just crossed 500,000M tokens burned! 🔥 People and companies are heavy users of open models on the Helmcode platform. To celebrate, we're sharing our opendata page, updated weekly. 👉 helmcode.com/opendata And a quick thread about companies usage 🧵

One model carries the load: Qwen 3.6 accounts for 60.7% of all tokens (348.8B). The surprise is the newcomer: GLM 5.2 is already at 5.9% (34.2B tokens) just with 2 weeks after we added it. (We are waiting for Kimi 3 open weights to choose between GLM 5.2 or Kimi3) Leaderboards tell you what benchmarks well. This tells you what people actually run. And Qwen 3.5 feels like Opus 4.6. There's no other model delivering that output at that price.

13.9B tokens in a single day (our first hackaton on the platform). That's our 90-day peak. Daily volume has gone from ~1B to a steady 8B+ in three months. Growth comes from companies shipping things that consume tokens every day. Open models for the win

40.6% of teams connect through the OpenAI SDK for Python. Add the JS SDK and it's over half of all traffic. That was the design goal: drop-in compatibility. Point your existing code at our endpoint, change one URL, done. @vercel AI SDK is moving up fast! (Honorable mention to the 12.9% still on curl. Respect.)