Cactus Needle 3: A Sliceable 8-29MB Automation Foundation Model That Matches DeepSeek v4 Flash
Cactus Compute introduced Cactus Needle 3, a new automation foundation model available in various sizes, including 29MB, 52MB, 98MB, and 121M. The Needle3-20L-121M model achieved an 86.0 score in Mobile Actions, outperforming LFM2.5 1.2B and Qwen3.5 0.8B. While smaller Needle3 subnetworks show weaker performance in BFCL and extraction tasks compared to larger baselines, the Needle3-20L-121M model demonstrates strong capabilities, matching DeepSeek v4 Flash in some metrics.
This report details how Cactus Needle 3, despite its small size, matches DeepSeek v4 Flash in some metrics, unlike other larger models that typically outperform smaller ones.
Time & source
Times shown in UTC
Display time zone: UTC
Local time zone unavailable; showing UTC.
IngestedOffset at this time: UTC+0Sep 18, 2026, 00:00 UTC
- Ingested
- Sep 18, 2026, 00:00
- Source type
- Dev community
Discussion trend
The percentage is based on collected discussion signal, not new comments or independent people. The curve only compares the same topic across time.
Full text isn't available here.
Read at source →