🚀 We release CLR, the test-time scaling method previously used in VibeThinker-3B.
CLR is built on a simple idea that solving and falsification have asymmetric capability requirements. With the same model, establishing a correct solution through forward reasoning is harder than falsifying one decisive wrong claim.
Rather than verifying entire solutions, CLR focuses on decision-critical claims and suppresses unreliable reasoning—compressing the survival space of incorrect traces and allowing more reliable ones to emerge.
🔥 On GPT-OSS-20B, when a correct trace is already present but standard self-consistency still fails, CLR rescues ~37% of such cases on average.
📄 Paper: huggingface.co/papers/2608.1…
💻 Code: github.com/WeiboAI/CLR
🎉 Accepted to the COLM'26 Workshop on Efficient Reasoning.
#LLM #TestTimeScaling #COLM #VibeThinker
Thanks @prz_chojecki for the independent evaluation.
Glad to see VibeThinker-3B performing strongly on IMO-like benchmarks.👏
We welcome more open, reasoning-focused tests from the community.
Thanks @stevibe for the independent test.🙏
Glad to see VibeThinker-3B performing strongly on this sliding-puzzle test. 🧠
We welcome more open, reasoning evaluations from the community.
A 3B model just cleared a puzzle that a 1.6 TRILLION param model couldn't.
You've seen this benchmark before: my sliding-puzzle test. Same Kimi & DeepSeek runs as last time. The only new thing: I dropped VibeThinker-3B in for a side-by-side.
> VibeThinker → 3B
> DeepSeek V4 Flash → 284B
> Kimi K2.6 → 1T
> DeepSeek V4 Pro → 1.6T
Shuffle depths 5, 10, 12, 15, 18, 22. One wrong move scrambles the whole board, so it's pure long-chain reasoning.
✅ VibeThinker-3B: solved all six. Never lost the thread.
⚠️ The giants started cracking at depth 15: Flash, Pro, and even Kimi each blew a run, scrambling the board past the move cap.
As VibeThinker was not trained for tool calling, I had it emit X and ran the move.
Bigger generalist ≠ smarter.
WeiboLLM retweeted
A 3B model just cleared a puzzle that a 1.6 TRILLION param model couldn't.
You've seen this benchmark before: my sliding-puzzle test. Same Kimi & DeepSeek runs as last time. The only new thing: I dropped VibeThinker-3B in for a side-by-side.
> VibeThinker → 3B
> DeepSeek V4 Flash → 284B
> Kimi K2.6 → 1T
> DeepSeek V4 Pro → 1.6T
Shuffle depths 5, 10, 12, 15, 18, 22. One wrong move scrambles the whole board, so it's pure long-chain reasoning.
✅ VibeThinker-3B: solved all six. Never lost the thread.
⚠️ The giants started cracking at depth 15: Flash, Pro, and even Kimi each blew a run, scrambling the board past the move cap.
As VibeThinker was not trained for tool calling, I had it emit X and ran the move.
Bigger generalist ≠ smarter.
⭐ VibeThinker-3B is released — a dense 3B model for frontier-level verifiable reasoning.
🚀 Reasoning: 94.3 on AIME’26, 76.4 on IMO-AnsBench, and 80.2 Pass@1 on LCB v6; with CLR, AIME‘26 improves to 97.1 and IMO-AnsBench to 80.6.
💻 OOD Coding: On recent unseen LeetCode weekly contests, VibeThinker-3B passes 123/128 (96.1%) first-attempt Python submissions.
⚡ Efficiency: Only 3B parameters, yet reaching the performance range of much larger top-tier reasoning models.
🧠 Perspective: Small models are not just cheaper substitutes. In parameter-dense domains with clear verification signals, SLMs offer a path to frontier-level reasoning that complements traditional Scaling Law.
Model : huggingface.co/WeiboAI/VibeT…
Github: github.com/WeiboAI/VibeThink…
Paper: huggingface.co/papers/2606.1…
#AI #LLM #Reasoning #OpenSource #SmallModel
VibeThinker-1.5B hit #1 on @huggingface ’s trending models today! 🔥
Huge thank you to our amazing community for the love, downloads, and priceless feedback.❤️
⭐ VibeThinker-1.5B — SOTA reasoning in a tiny model.
🚀 Performance: Highly competitive on AIME24/25 & HMMT25 — surpasses DeepSeek R1-0120 on math, and outperforms same-size models in competitive coding.
⚡ Efficiency: Only 1.5B params — 100-600× smaller than giants like Kimi K2 & DeepSeek R1.
💰 Cost: Full post-training for just $7.8K — 30-60× cheaper than DeepSeek R1 or MiniMax-M1.
🧠 Innovation: Powered by our Spectrum-to-Signal Principle (SSP) and MGPO framework.
Model : huggingface.co/WeiboAI/VibeT…
Github: github.com/WeiboAI/VibeThink…
Arxiv : arxiv.org/abs/2511.06221
#AI #LLM #Reasoning #OpenSource #SmallModel
