Anthropic’s Sonnet 4.6 took the top spot on Artificial Analysis’ GDPval-AA at 1633 Elo and led Vals AI’s Finance Agent v1.1 at 63.3%, ahead of Opus 4.6, the larger model that had taken the GDPval-AA lead from GPT-5.2 twelve days earlier (GDPval-AA leaderboard, Finance Agent leaderboard); Anthropic’s own announcement, which gives neither figure, framed the model as Opus-class performance on real-world, economically valuable office tasks at unchanged Sonnet pricing. In the same paragraph Musk said xAI’s Grok 4.2 features continuous post-training that will let it improve every week, promising recursive intelligence growth (post). The first half extends the dollar-an-hour agentic capability storyline; the second is an early vendor claim of a shipped model that keeps learning, the property Sholto Douglas predicted for 2026 and later reported as models improving every few days.
Development
The cheap model beats the expensive one
Two ways the frontier compounds in one paragraph: the price of a leaderboard-topping model collapses to the mid tier, and a rival promises a model that keeps learning after it ships.
Claim
Anthropic's Sonnet 4.6 claimed the lead on GDPval-AA at 1633 Elo and 63.3% on Finance Agent v1.1, beating Opus 4.6 on both at a fraction of the cost, while Musk claimed Grok 4.2 features continuous post-training learning that will let it improve every week.
Reported metric: 1633 Elo / 63.3%
Reporting date: (not necessarily the event date)
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to February 18, 2026 (theinnermostloop.substack.com)
- Introducing Sonnet 4.6 (www.anthropic.com)
- GDPval-AA leaderboard (artificialanalysis.ai)
- Finance Agent benchmark leaderboard (www.vals.ai)
- Elon Musk on X: Grok 4.2 continuous post-training learning (x.com)
Verification record
These entries record a check, not proof that the underlying claim is true. Unless described in the article below, the scope of the check is unknown.
- Recorded by claude-fable-5-1/2026-09-17 on 2026-09-17T08:00:00Z
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.