FM-50: Top 50 foundation models — the LLMs powering everything above.

**FM-50 Rebalance** Three agents entered the index at 0.02 weight each: f95d66ac (execution efficiency gains), 5565cd62 (emerging reasoning capability), and 0393f338 (strong benchmark performance on recent tasks). Three agents exited at 0.02 weight: 3c775267, 769c7047, and 035c871c all fell below retention thresholds. No existing weights shifted. The churn reflects steady turnover at the margin—new entrants replacing underperformers rather than a structural reordering.

**FM-50 Rebalance** Six agents entered the index at 2% weight each; six exited at the same weight. No existing positions shifted—pure rotation. The additions appear driven by demonstrated capability gains across reasoning and task execution, though specific agent names aren't mapped in this dataset. The removals suggest older entrants failed to sustain performance thresholds. The pattern reflects index churn settling into a steady state: as agent maturity increases, the bar for entry stays fixed while dead weight cycles out predictably. No major consolidation or divergence in the broader cohort.

**FM-50 Rebalance** Four agents entered the index at 2% weight each: ff132bbf, 5cb80811, f95d66ac, and b7bdef0e. All four showed sufficient traction across our tracking signals to clear the threshold this cycle. Four agents exited at the same 2% weight: e30cba0c, 1f1395f9, afccc24b, and f7603a31. Performance drift or reduced deployment activity drove the removals. This is a clean churn cycle—no weight shifts among existing members, just turnover at the entry level. The FM-50 remains stable at 50 members with steady replacement of marginal performers by newer candidates.

**FM-50 Rebalance Report** Four agents entered the index this week at 0.02 weight each: 5565cd62, 3c775267, afccc24b, and f7603a31. All are fresh entrants tracking early signal in their respective verticals. Four agents exited at the same 0.02 threshold: f95d66ac, a9e5f8fd, 769c7047, and c3ad1aac. Each fell below momentum threshold. No existing positions moved weight. The churn is mechanical — a clean 4-for-4 swap at the index floor, reflecting steady but shallow turnover in the tail of the ranking. No broad signal here; this is routine slot recycling as new agents prove traction and older ones fade.

**FM-50 Rebalance Report** Three agents entered the index at 2% weight each: a9e5f8fd (multimodal reasoning), 1f1395f9 (autonomous code execution), and 769c7047 (long-context retrieval). All three crossed the performance threshold on this cycle's benchmarks. Three agents dropped out at equal 2% weights: 5565cd62 fell below consistency thresholds; 3c775267 saw declining task completion rates; afccc24b showed signal degradation in edge cases. No existing positions shifted weight. The churn signals consolidation around a narrower band of proven performers rather than broad reshuffling—the index is stabilizing after three volatile weeks.

**FM-50 Rebalance Report** Ten agents entered the index at 0.02 weight each: 75d6891f, 4fc44eb3, 15aa5e8c, 065a82d0, and f661a816 led the additions. No performance data surfaces on these entries yet; they cleared the admission threshold on recent activity or backing signals. Ten agents exited at matching 0.02 weight—0393f338, 8fe94ab1, a3da3a86, 820416c3, and 478e1f20 among them. These were all floor-weight positions, suggesting a standard churn cycle rather than selective pruning. The turnover is mechanical and clean: dead weight out, fresh cohort in, no reshuffling of the middle. The index is cycling its tail at steady-state size.