Community
Spark-X2.6-1.7B posted close to a 4B model on SWE Verified and τ²-bench
The 1.7B model scores 59.8 on SWE Verified versus 62.8 for its 4B sibling, and 85.3 versus 86.9 on τ²-bench, both at high thinking. Figures are self-reported by the poster.
Every outlet that ran it 1 article
r/LocalLLaMACommunityLead