Vals AI posted on X at 12:04 Beijing time on October 20, 2026, that the Vals Index had moved to v2.2. The Vals Index page is marked Updated 10/19/2026 and lists the same top three and open-weights leader. The index aggregates agentic work across finance, coding, legal and tax, weighted by each sector's share of U.S. GDP, according to Vals AI.
What changed in v2.2
According to the Vals AI X thread and the Index updates log, v2.2 includes:
- Finance Agent v2.1: a new panel of judge models, and answers that hedge between values now fail
- Excel Modeling Benchmark v1.1: graded by GPT-6 Luna
- Code Migration v1.1: every model runs on the Pi agent harness, with revised hidden tests
- Harvey's LAB v1.1 (HLAB on the leaderboard): judges can see tracked changes in redlined documents
- Tax Agent Bench: a question counts only if every rubric check passes (all-pass accuracy)
New top of the board
On the Vals Index leaderboard as of the October 19, 2026 update, Vals AI reports:
- 1. Claude Sonnet 5.5 — 70.85%
- 2. Claude Opus 5.5 — 70.18%
- 3. Gemini 4 Argon — 69.24%
- 8. DeepSeek V4.1 Flash 1017 — 61.26%, described by Vals AI as the top open-weights model
Vals AI notes that Sonnet 5.5 and Opus 5.5 scores include some component tasks served by a fallback model after a provider refusal; counting those tasks as failures, Vals AI says, would drop Sonnet 5.5 to 69.03% and Opus 5.5 to 64.16%, while Gemini 4 Argon had no fallbacks and would rank first. That condition is the key caveat on this ranking. Component leadership still splits: Vals AI says Opus 5.5 leads Finance Agent v2.1, while Sonnet 5.5 leads EMB v1.1 and Code Migration v1.1, and Gemini 4 Argon leads HLAB v1.1, among other splits on the page.
