Launch weights are derived from our review-page benchmark table. Any change here must remain consistent with a published review result — if the arena and the review disagree, the arena bends.
Arena v0.1 — Launch Synergies
Initial synergy weights. Claude Code + Opus leads at +15 Precision. Codex + GPT-5 at +12. FAISS + LlamaIndex unlocks EMP.
Diff
claude-opus-4-7 + claude-code
precision +15
Co-designed harness, thinking-stream pass-through. Derived from reference benchmarks on long-horizon coding tasks.
gpt-5-core + codex-cli
precision +12
Codex's parallel-agent pattern matches GPT-5's structured-output throughput.
index-faiss + rag-llamaindex
synergy +1
Pairing an Index with a RAG layer unlocks the EMP weapon — gate Precision ≥ 75 with both.