Mercor @mercor

I couldn’t be more excited to work with @edwardjhu As the creator of LoRA (fine-tuning) and a core contributor to o1 at OpenAI, few people have Edward’s combination of technical depth and vision. Our research team is shaping how human intelligence will power the AI economy, which is the most important problem of the decade. More soon 🚀

Edward Hu @edwardjhu

8 hours ago

I am joining Mercor to lead model training & research. We envision a world with an abundance of intelligence. Great data is increasingly the bottleneck for frontier models to tackle economically valuable work. We believe great data can is best produced in connection with great

keyboard_arrow_left Previous keyboard_arrow_right Next

26 8 323 24K 99

Post-training is becoming more accessible by the day. We're open-sourcing our post-training recipes and the end-to-end process. Every enterprise will soon post-train its own models.

Edward Hu @edwardjhu

8 hours ago

How does one RL post-train a 397B model for long-horizon knowledge work? 👩‍💼 We share every step we took to bring Qwen 3.5 397B from 16.1% Pass@1 to 27.3% on APEX-Agents using DPPO, including final models weights and the full training script🚀 This is the first of many works from

8 50 376 58K 455

Data is the most important ingredient in post-training. The Mercor Research team focuses on making every hour of expert work yield the most model improvement, through better learning algorithms for knowledge work and automated, domain-specific post-training. We're also committed to doing open source research. That's why we're publishing a RL training guide for Qwen3.5-397B in collaboration with the SkyRL team. Find out how we raised Pass@1 on APEX-Agents from 16% to 27% and explore the full training script, model weights, and eval traces. Want to do this kind of work with us? We're hiring. Read the full blog post: mercor.com/blog/training-…

Fable 5.1 increases its scores on both APEX-SWE domains. Integration: 63.5% -> 68.1% (+4.6) Observability: 54.2% -> 59.0% (+4.8) Even with these gains, Fable 5.1 failed 59 of the 199 tasks outright.

Fable 5.1 debuts at #2 on the APEX-SWE leaderboard, within the confidence band of first place. It’s also the new leader for Integration tasks. Overall: 63.6% (#2) Integration: 68.1% (#1) Observability: 59.0% (#2) Congratulations to @AnthropicAI @claudeai.

See the full leaderboard: mercor.com/apex/apex-agen…

Fable 5.1 solves 48 tasks that Fable 5 could not. This suggests the model has improved its ability to understand and reason through complex concepts within professional services. Here is the distribution of the failure modes 5.1 improved on that lead to successful completion of those tasks: Reasoning: 31 Planning and reflection: 9 Information gathering: 6 Instruction-following: 2

Fable 5.1 sweeps all three APEX-Agents domains on Pass@1. 📊Management consulting: 52.2% (#1) 💰Investment banking: 48.3% (#1) 🏛️Corporate law: 41.9% (#1) The model demonstrates significant capability for agentic coding, long-running agentic workflows, and knowledge work. Congratulations to @AnthropicAI @claudeai.

An open letter for a global surge in cyber defense, signed by over 100 organizations including Anthropic, AWS, Google, Microsoft, OpenAI, and Oracle. x.com/i/article/2093…

Today we're introducing the Mercor Research Fellowship. We're funding a small group of people to build a new APEX benchmark, the definitive measure of frontier AI on professional-quality work. Applications are now open: mercor.com/careers/?ashby…

APEX-Agents is one of the deepest benchmarks in legal work. @mercor put a ton of effort into realistic worlds with large filesystems and details down to calendars, inboxes and more all generated by human experts actually solving tasks. That gains on synthetic data like LAB generalize to scores on this benchmark was one of our most promising results and its been awesome to partner with Mercor on both benchmarks (and some very cool future ones!)

Mercor @mercor

2 weeks ago

We’re proud to collaborate with @harvey on their Tenet model. Harvey evaluated Tenet on APEX-Agents Corporate Law, with tasks authored by practicing attorneys. Post-training Kimi K3 with @FireworksAI_HQ, Tenet was able to achieve +15.2 points lift over the base model. When

7 10 87 14K 18

We’re proud to collaborate with @harvey on their Tenet model. Harvey evaluated Tenet on APEX-Agents Corporate Law, with tasks authored by practicing attorneys. Post-training Kimi K3 with @FireworksAI_HQ, Tenet was able to achieve +15.2 points lift over the base model. When compared to generalist frontier models, Harvey ranks #1 on APEX-Agents in the Corporate Law domain with 74.0% mean score. Harvey is leading the wave of application-layer AI companies to own their intelligence with their Tenet series of models. See the full leaderboard: mercor.com/apex/apex-agen…

Harvey is leading the wave of application-layer AI companies to own their intelligence with the Tenet series of models. They're starting with a foundation model for law, then expanding to custom models for each customer. This is the leading indicator for every Enterprise. They will post-train an open-source model for their domain (accounting, sales, design, etc), then customize those models for every major customer. @mercor is proud to partner with @harvey to make this possible.

Harvey @harvey

2 weeks ago

Introducing Tenet, our first model post-trained for legal. Tenet is a Kimi K3 base that we post-trained with @FireworksAI_HQ on a corpus of publicly available legal data, synthetic data, and human expert data simulating long-horizon legal work. Training increases Tenet's

122 281 3K 3.5M 3K

Congratulations to the @Harvey team on releasing their Tenet model. Human expert data is crucial for post-training work. We’re proud to have collaborated with @calvincongelado @vtrengarajan @ItsJulioPereyra @nikogrupen @gabepereyra, the Harvey research team, and hundreds of Mercor experts on this project.

Harvey @harvey

2 weeks ago

Introducing Tenet, our first model post-trained for legal. Tenet is a Kimi K3 base that we post-trained with @FireworksAI_HQ on a corpus of publicly available legal data, synthetic data, and human expert data simulating long-horizon legal work. Training increases Tenet's

122 281 3K 3.5M 3K

Great article by @sonyatweetybird, esp on the importance of high-quality evals/benchmarks. "Most evals start as a founder squinting at outputs and vibe-checking whether they feel right." ^we started here but have since invested an enormous amount of time, effort, and resources into building Legal Agent Bench and other internal eval sets to turn legal judgment into a hill-climable signal. It's created a powerful flywheel for us @harvey. h/t @ItsJulioPereyra's team of legal researchers who bring their much needed domain expertise to the equation and at @BrendanFoody @mercor for helping us scale it up.

Sonya Huang 🐥 @sonyatweetybird

2 weeks ago

The race for the AI application layer is not only about UI, workflows, or GTM... it is a fight for the intelligence layer itself. x.com/i/article/2088…

38 37 354 74K 674

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. @mercor’s @BrendanFoody joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?

Grok 4.6 is one of the most cost efficient frontier models we’ve ever tested on APEX.

Elon Musk @elonmusk

3 weeks ago

Grok 4.6 is now out 🚀🚀🚀 Smart, fast & amazing bang for buck!

3K 3K 25K 7.9M 1K

Compare results on the full leaderboard: mercor.com/apex/

The two APEX-SWE task types improved slightly over Grok 4.5. Integration, which measures building and deploying across services climbed from 60.3% to 62.7%. Observability, which tests debugging from production telemetry increased from 47.0% to 50.1%.

Grok 4.6 is in the top four across three APEX productivity benchmarks. APEX-Agents: 57.5% mean score, #4 overall APEX-Accounting: 50.9% mean score, #4 overall APEX-SWE: 56.4% Pass@1, #3 overall Congratulations to @SpaceXAI.