Organizing human intelligence is the most important problem in society.
Every enterprise will need humans to constantly refine their system of record for evaluating and specifying agent behavior.
As AI becomes more powerful, humans will become more important not less.
Mercor
291 posts
Organizing human intelligence to power the AI economy.
- Mercor repostedA year ago, I predicted that we would enter The Era of Evals, but it's now happening much faster than I anticipated. Frontier Labs have scaled their Eval production with us by more than 10X in the last 12 months, on what was already a 9-figure base. Every tech-forwardMercor (@mercor_ai) is now working with 6 out of the Magnificent 7, all of the top 5 AI labs, and most of the top application layer companies. One trend is common across every customer: we are entering The Era of Evals. RL is becoming so effective that models will be able to
- Mercor repostedTwo days left to apply for the Inference-Time Compute Hackathon, hosted by Cognition, @mercor_ai, @Etched, and @AnthropicAI, where you get: • 8x H100s per team • $100k+ in prizes • Dedicated Agents Track Build something that pushes the frontier.
- Agents are only as good as the environments behind them. At Mercor, we've built deep expertise in the realistic, economically-grounded environments that help agents bridge the gap from the lab to real-world usefulness. We want to put that expertise to work for the broader
- Mercor repostedClaude Fable 5's progress in coding (APEX SWE) dramatically outpaced progress in other domains like finance, law, and consulting (APEX Agents). The top reason for the fast progress in coding is that we have GitHub, with over 28 million repositories of human-written code. We're
- Claude Fable 5 places 2nd on APEX-Agents leaderboard @claudeai Fable 5 (Max) scores 45.0% Pass@1, behind Gemini 3.5 Flash (49.6%) and ahead of Claude Opus 4.8 (42.5%). Fable 5 reached 2nd overall while spending far fewer tokens. It used 70% less than Gemini 3.5 Flash and 37%APEX-Agents domain breakdown for Claude Fable 5 (Max), Pass@1: Corporate Law: 40.9% (1st) Investment Banking: 47.7% (2nd) Management Consulting: 46.4% (2nd) With 4 runs, Fable solved 246 of 480 tasks, including 8 that no other model has solved. All 8 tasks are in Law,Sign up for the APEX-Agents newsletter: mercor.com/apex Download the APEX-Agents dataset: huggingface.co/datasets/merco… Open-source infra + eval service (Archipelago): github.com/Mercor-Intelli… Technical report: arxiv.org/abs/2601.14242
- Mercor repostedWhat are the top AI researchers saying about the future of human work? @BrendanFoody, co-founder and CEO of Mercor: "People often have this misconception that we won't need data in three years because we'll have super intelligence... AI better than humans at absolutely00:00Mercor's average pay rate just surpassed $100 / hour. We're mobilizing tens of thousands of top software engineers, bankers, lawyers, and doctors to build the next generation of AI models. Every top economist believes that there will be more jobs in 10 years than there are
- Mercor repostedClaude Fable 5's progress on hillclimbing APEX-SWE is accelerating exponentially. While other models focus on reasoning over a code base, Claude has unparalleled results at reasoning over Linear tickets, observability logs, Slack messages, and Google Drive files alongside theClaude Fable 5 takes #1 on APEX-SWE: 65.5% Pass@1 overall. It scores ~18pp higher than Opus 4.8. We tested @claudeai Fable 5 on APEX-SWE which measures whether AI models can do real software engineering work. Fable 5 tops our two APEX-SWE categories: - Integration: 61.3% -
- Claude Fable 5 takes #1 on APEX-SWE: 65.5% Pass@1 overall. It scores ~18pp higher than Opus 4.8. We tested @claudeai Fable 5 on APEX-SWE which measures whether AI models can do real software engineering work. Fable 5 tops our two APEX-SWE categories: - Integration: 61.3% -What changed is how Fable 5 works. Where Opus 4.8 reads code one file at a time, Fable 5 investigates in parallel: 10.4 parallel tool calls per trajectory versus zero for Opus 4.8. It searches code, reads files, runs tests, and queries logs at once, instead of chasing oneThe pattern is higher leverage, not more effort. Fable 5 spends less of its budget searching and more validating, testing as it goes and catching issues during edits rather than after. It rewrites less too: 3.8 edit iterations versus 6.0 for Opus 4.8. The result: 10% fewer
- Mercor repostedAI progress requires (1) compute, (2) algorithms, and (3) data. - The leading compute company is worth $5 trillion. - The leading model company is worth $1 trillion. - @mercor_ai is the leading data company and is currently valued orders of magnitude lower. There's an
- Mercor repostedI told @HarryStebbings on @20vcFund that @mercor_ai now spends more on tokens for our internal agents than we do on headcount. In a few years, every enterprise will too. We already see that the companies pulling ahead are running like AI labs: an eval for every workflow, agents
- We worked with @trajectorylabs to run their SDPO++ algorithm on APEX-Agents and see what it could do with real production data. Pass rates went from 5% to 25% on GPT-OSS-120B, and the curve is still climbing. Read more about our work together in their blog post below.5 Days of Trajectory 🏹Day 5: Scaling SDPO to Agentic Tasks Continual learning means you must train on data from production. But production gives you one example per task. A user makes a request once. You get one trajectory, not a batch. However, current RL algorithms
