rsi.watch

Apex Intelligence · 2026-10-28

Introducing Apex-35B-V1.0

“We built Apex Research, an automated research system that enables recursive self-improvement (RSI) by continuously identifying weaknesses, generating hypotheses, running experiments, and turning discoveries into measurable model improvements.”
Who
Apex Intelligence
Date
2026-10-28 The post shows no date; the Hugging Face record for the model repository gives 2026-10-28.
Evidence
Not rated. A description of how the model was built. The benchmark figures measure the model. The post reports CFD-Bench as wins, draws and losses against Qwen3.6-35B-A3B, and gives no baseline or measurement setup for the 32.1-point, throughput or decoding figures it attributes to Apex Research. how we rate evidence
Replication
Nothing to rerun. all claims
Checked
2026-10-29, against the source page
Links
Source

Also on the page

  • The model is "based on Qwen3.6-35B-A3B and has been further developed through mid-training and post-training". Weights are on Hugging Face.
  • Apex's definition: "By “self-evolving,” we mean recursive self-improvement (RSI): enabling AI to participate in research, experimentation, and validation, then using the resulting discoveries to improve the model itself."
  • Figures Apex attributes to Apex Research: a self-evolving data pipeline that "improved performance on our internal CFD-Bench by 32.1 points", "45–60% higher training throughput", and "1.6–1.8× faster decoding" "under comparable workloads".
  • Apex's post compares the model with Recursive Flint and says it outperforms Flint and Qwen3.6-35B-A3B "across all six research benchmarks". Apex ran those models itself, and "Every task is run once (pass@1)."
  • Apex was founded by Yongchao Chen, "an assistant professor at Tsinghua University’s College of AI".

Related claims

← Ledger Report an error