Uploaded September 2025 | Updated September 2026, 1 hour ago
🚨 ByteDance just dropped an open-weight 36B parameter model with a 512,000-token context window—that’s the longest native context you can get in an open model today.
They call it Seed-OSS-36B, and it’s not just about size. The model introduces a thinking budget you can dial up or down:
0 tokens = instant direct reply
512 tokens = quick Q&A
2k–8k tokens = deeper reasoning, coding, or tool use
It even self-reflects mid-generation to stay efficient. On top of that, it ships under Apache 2.0, meaning you can use it commercially.
There are three flavors:
Instruct (for agents, coding, reasoning, long docs)
Base + synthetic data (for evaluations with strong performance)
Base clean (for research without synthetic bias)
In short: Bytedance just made long-context + controllable reasoning available to everyone. Expect a lot of innovation built on top of this.
Which release should I break down next? Drop it in the comments and I’ll cover it.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AInews #LLM #ArtificialIntelligence #short
🚨 ByteDance just dropped an open-weight 36B parameter model with a 512,000-token context window—that’s the longest native context you can get in an open model today.
They call it Seed-OSS-36B, and it’s not just about size. The model introduces a thinking budget you can dial up or down:
0 tokens = instant direct reply
512 tokens = quick Q&A
2k–8k tokens = deeper reasoning, coding, or tool use
It even self-reflects mid-generation to stay efficient. On top of that, it ships under Apache 2.0, meaning you can use it commercially.
There are three flavors:
Instruct (for agents, coding, reasoning, long docs)
Base + synthetic data (for evaluations with strong performance)
Base clean (for research without synthetic bias)
In short: Bytedance just made long-context + controllable reasoning available to everyone. Expect a lot of innovation built on top of this.
Which release should I break down next? Drop it in the comments and I’ll cover it.
I’m Louis-François, PhD dropout, now CTO & co-founder at Towards AI. Follow me for tomorrow’s no-BS AI roundup 🚀
#AInews #LLM #ArtificialIntelligence #short
