Open-weight Qwen3.5-397B plus continual learning reportedly reaches Opus 4.8-level performance for $450k
WeijiaShi2 · x · 2026-07-28
- A thread claims that applying π-shaped continual learning to an open-weight model like Qwen3.5-397B can produce a “genuine frontier model” competitive with Opus 4.8 for about $450k in compute.
- The post also says a newly curated dataset behind Thomson-1-Large focuses on legal capabilities and reportedly matches Opus-4.8 on general benchmarks.
- The author says a full technical report and open-source models are coming soon.
Related event: SovereignAI Boosts Qwen to Near-Opus for $450K(4 posts)→
More from Companies & People
- Indie dev: after fighting Apple review for a month, every new release brings PTSD — jdluk87 · 2026-09-23
- PrimeIntellect Adds Intern to Optimize Its Inference Stack — willcb · 2026-09-23
- Why OpenAI's vertical push into law and finance is 'dead on arrival' — nicolechirps · 2026-09-23
- Cloudflare CTO Dane Knecht makes TIME's 2026 executives list as AI crawlers hit 52% of traffic — dinasaur_404 · 2026-09-23
- tszzl: glossy 3D animation is now the single most important skill for launching a new AI model — tszzl · 2026-09-23
- Sam Altman on his highest-return year: dozens of textbooks, helping people, planting seeds — a16z · 2026-09-23