Thinking Machines Lab's Inkling just got its first independent stress test on genuinely hard knowledge work, and the results are a mixed bag in the most instructive way. Artificial Analysis ran the model through AA-Briefcase, its agentic knowledge-work benchmark, and Inkling landed at an Elo of 836 -- ahead of DeepSeek V4 Flash but below the leading open-weights models, GLM-5.2 and Nemotron 3 Ultra.

What AA-Briefcase actually tests

Most benchmarks ask a model a question and check the answer. AA-Briefcase is different. It evaluates models across four multi-week knowledge work projects, comprising thousands of input files and 91 tasks in total, spanning fields like data science, product management, and corporate strategy. The output isn't a text answer -- it's a deliverable: a spreadsheet, a PowerPoint deck, a PDF memo, or an HTML mock-up.

Each task is graded on three independent dimensions:

  • Rubric checks: Binary pass/fail. Did the model follow instructions, find the right evidence across source files, and reach correct conclusions?
  • Analytical Quality: Pairwise comparison against another model's submission. Which one is more thorough and better-structured?
  • Presentation Quality: Another pairwise comparison. Which deliverable looks more professionally done?

The AA-Briefcase Elo is a single metric that combines results across all three dimensions. It's a harder and more realistic signal than most leaderboards, because it's measuring whether an agent can carry a complex knowledge-work project through successive stages and produce accurate, complete, well-presented deliverables.

Where Inkling lands on the leaderboard

The full leaderboard tells a clear story about where the frontier sits. Claude Fable 5 leads at 1587 Elo, with Claude Opus 4.8 and GLM-5.2 in second and third. Inkling at 836 sits in the middle of a crowded pack, roughly tied with Gemini 3.5 Flash and MiMo-V2.5-Pro.

Alpha Signal

Don't miss what's next in AI

Join 300,000+ engineers and researchers who get the signal, not the noise.

  • Full access to in-depth AI research breakdowns
  • Be the first to know what's trending before it hits mainstream
  • Daily curated papers, repos, and industry moves