Samaya AI

We are reaching an important inflection point in AI for finance: frontier models can now outperform expert analysts at predicting earnings surprises. A couple of months ago, we started a research effort at Samaya to study AI's predictive capabilities in finance. Every quarter, companies report key financial metrics during earnings that inform investment decisions. An enormous amount of analyst time goes into modeling, predicting, and publishing these metrics ahead of earnings. We gave AI agents the same task. We built a finance-specific prediction harness and environment to give AI models comprehensive, point-in-time financial information at parity with human experts so that we could push their capabilities to the limit. We found that the most recent capable models (Fable 5.1; Opus 5.5; GPT-6 Astra) are able to outperform the harder, bias-corrected consensus baseline. The models are able to calculate the financial impact of world events and news the way we expect from a strong human analyst. Anecdotally, the models’ ability to gather new evidence and willingness to adjust their view away from the consensus drive their wins over consensus. We have an alpha version of the earnings prediction agent within the Samaya product available for users and clients. We’re also working with clients on adapting these predictions to internal data to generate firm-specific insights. Get in touch to try it out and collaborate! 🔗 Full research in comments.

AI Agents used for investment decision making often fail on the details, not on the reasoning. A research agent can have the right tools and the right data and still miss that a company reports product lines only as a share of sales, or that a quarterly series has to be rebuilt from year-to-date filings. Our AI research team trained an advisor model to write a short, question-specific plan for our research agent using Hindsight Structured Advice Distillation (HSAD). The advisor learns from the agent's own history: we turn the agent's past failures into a checklist of recurring fixes, try several strategies to see which one worked, and with hindsight learn the plan that would have helped most. On FrontierFinance, our public benchmark of 220 expert-written research queries: 🔹 Our quality score rises from 52.9% to 56.6% of rubric items satisfied, the top score on the leaderboard 🔹 Above Samaya’s previously best high-effort configuration, at 37% lower cost per query 🔹 Above the best frontier model on the leaderboard (Claude Fable 5.1, 55.9%) 🔗 Blog in comments.

Daloopa is now live in Samaya AI. The 30 minutes after a print are when analysts need reported numbers most, and when most data feeds are still catching up. Daloopa tags as-reported data points within minutes of a filing, down to segments and KPIs, and links each one to the exact page it came from. If you already have a Daloopa license, you can now connect it to Samaya. When JPM reports, Daloopa has the financials tagged in minutes, and Samaya can build an actuals vs. consensus variance table, flag material beats/misses and drivers for the stock, and surface the questions you should be asking on the call. You’ll know how much net interest margin beat by, what shifted on charge-off rates, and how investment banking fees grew in less than 10 minutes, so you can spend more time diving deeper for what really matters. Try the platform purpose-built for your investment workflows, with Samaya quality across financials, consensus, live transcripts, broker research, news - and now Daloopa. To connect: open Samaya, go to Connectors, select Daloopa, and sign in with your existing Daloopa credentials.

50 new workflows are live on Samaya! Getting good output from AI starts with knowing what to ask. Samaya provides robust workflows right out of the box so you spend less time prompting. Each one is built around a single task and answers the questions specific to that output. Post-Earnings Call Takeaways takes the call and the release together and gives you the themes, what moved against the guide, how management handled the questions, and the numbers on the call that aren’t in the press release. Meeting Prep looks for new disclosures from management, the key themes you need to prepare for, and current events that can change the story for the company. Samaya sets you up with relevant questions so that your bases are covered when you’re at a conference with a nonstop meeting schedule. Those are just two of the 50 new workflows on Samaya, covering buy-side and sell-side work. The other forty-eight cover earnings prep, coverage monitoring, ramping on a new name, investor communications, and research reports. Each Samaya workflow is built by analysts who used to do this by hand. Which one will you try first?

Excited to be an early launch partner of Claude Fable 5.1! We partnered with Anthropic to evaluate Fable 5.1 on FrontierFinance, our hard benchmark of agentic investment workflows. We see a substantial gain in performance, with Claude Fable 5.1 scoring 55.9% compared to 49.2% for Fable 5 (the prior best performing frontier model.) This gain largely comes from Fable 5.1 carrying out more and better tool calls. The more tool calls does come with a cost increase however — Fable 5.1 is about 1.7x as expensive as Fable 5. Looking forward, effective and increased tool use will be a core part of powerful long horizon Agents. But with increased tool use comes increasing costs, and there's real work to optimize this so that Agents can use tools both effectively and efficiently.

🌟 Welcome to Samaya AI, Lomond Davis! Lomond started his career at Citi on the trading side before moving into equity research sales, and later joined YipitData in the world of alternative data. Across those roles, he kept coming back to the same challenge: when there’s more information than anyone can reasonably process, how do you find what actually matters? In equity research sales, Lomond describes his role as being the “human synthesis layer,” reading across research, understanding what was in each client’s book, and connecting them with the insights most relevant to their decisions. As the volume and complexity of available information has grown, he sees AI as an opportunity to tackle that challenge in an entirely new way. That’s what brought him to Samaya AI. As an Enterprise Account Executive, Lomond is excited to work closely with investors to understand where deeper analysis can improve their research process, and bring those insights back to the people building our technology. Outside of work, Lomond enjoys cooking, stand-up comedy, and electronic music. He’s also a part-owner of a kiwi sanctuary in New Zealand. 🥝

Thank you, NVIDIA for Startups, for highlighting Maithra Raghu and the vision behind Samaya AI. That ambition continues to shape what we’re building today: Expert AI Agents that turn real-time information into investment conviction with speed, precision, and accuracy. Proud to see that work recognized!

A month ago, we released FrontierFinance, an open benchmark testing the frontier capabilities of finance AI agents. The response from the community has been terrific, and we've already seen encouraging progress from the latest models on these tasks. Today, we're publishing the FrontierFinance technical report. In the report, we detail our annotation process, share an in-depth analysis of the dataset, and analyze into how current systems and models actually behave, surfacing insights we hope will inform how future frontier models and harnesses are built. Read the full report via the link in the comments. 👇

Imagine an alert that understands that a document has referenced something you care about. Set up a content-based alert in Samaya and get notified when an expert transcript about a competitor you don’t follow mentions a company where you’re holding a long position. Try: “Monitor expert transcripts for mentions of ad tech that impact companies like Meta and Google.” Set an alert and let Samaya surface the read-acrosses you might otherwise miss.

Bond data is live in Samaya AI, through Finnhub. Historical and last price, maturity, coupon, and bond details across US corporate issuers. Ask for every outstanding issuance on a name and get the full stack back, with last price and yield on each line, cited to the source. Credit and equity analysts have been working the same issuer from two different tools. Now the capital structure sits next to the filings, the transcripts, and the model. Available to all Samaya accounts today.