To train better earnings forecasters, we need events whose outcomes the model hasn’t already learned. But how far back can we reliably evaluate AI earnings predictions without prior knowledge affecting the results? When prompted directly, Nemotron Ultra recalled exact EPS for just 2 of 272 historical events. Yet its answer probabilities showed a different pattern: the oldest earnings answers received ~75x the probability of the most recent ones. Recall can miss prior-knowledge signals. Measuring them could help us choose more usable post-training data. Read the research ↓

New in Samaya: Memory. Tell Samaya you want margin moves in basis points and comparisons against your five names. Three weeks later, ask about Q3 earnings. Samaya remembers your preferences and applies them automatically. Your research assistant learns how you work.

We gave AI agents the same earnings prediction task as expert analysts. Using Samaya’s finance-specific prediction harness, the latest frontier models (Fable 5.1, Opus 5.5, GPT-6 Astra) outperformed a harder, bias-corrected consensus baseline. The models gathered new evidence, calculated the financial impact of events, and adjusted their predictions away from consensus. This is a critical inflection point for AI in finance. Full research ↓

AI research agents often fail on the details, not the reasoning. We trained an advisor model using Hindsight Structured Advice Distillation (HSAD) to learn from our research agent’s past failures and guide it with a question-specific plan. The result: better research quality without spending more compute. On FrontierFinance, our public benchmark: 🔹 56.6% quality, #1 on the leaderboard 🔹 Beats our previous best at 37% lower cost 🔹 Outperforms the best frontier model (Claude Fable 5.1, 55.9%) More on how it works ↓

Fifty new workflows are live in Samaya. A hard-to-get management meeting just landed on your calendar. The last thing you want is to waste that time asking a question they've already answered. Run a Meeting Prep workflow to catch up on the key debates, current events, what management said at the last conference, and the questions you need answered, so you can walk into the room prepared. Buy-side and sell-side. Built by analysts who used to do it by hand.

A month ago we released FrontierFinance, an open benchmark for the frontier capabilities of finance AI agents. The response from community has been terrific, and the latest models are already making fast progress. Today we're publishing the technical report: our annotation process, a deep analysis of the dataset, and a close look at how current systems and models actually behave, with insights we hope will shape the next generation of frontier models and harnesses. Full report 👇

New results in: Gemini 3.6 Flash achieves 46.3% on FrontierFinance, our open benchmark for finance AI agents. Ahead of Claude Opus 4.8, on par with GPT-5.6 Sol, and a big jump over past Gemini generations. What stands out is the efficiency: it hits that score at just $2.41 per query, cheaper than both, and at 164s it's among the fastest models we've tested on this harness. In our analysis, it's notably stronger than Opus 4.8 at surfacing qualitative and contextual insight, and leads on Screening & Discovery, one of the benchmark's hardest use cases. More about FrontierFinance and the full benchmarking results 👇

𝗢𝘂𝗿 𝗻𝗲𝘄 𝗵𝗶𝗴𝗵𝗲𝗿-𝗲𝗳𝗳𝗼𝗿𝘁 𝘃𝗲𝗿𝘀𝗶𝗼𝗻 𝗼𝗳 𝗦𝗮𝗺𝗮𝘆𝗮 𝗮𝗰𝗵𝗶𝗲𝘃𝗲𝘀 𝟱𝟲% 𝗮𝗰𝗰𝘂𝗿𝗮𝗰𝘆, 𝗮 𝗻𝗲𝘄 𝘀𝘁𝗮𝘁𝗲 𝗼𝗳 𝘁𝗵𝗲 𝗮𝗿𝘁, 𝗮𝘁 𝟮𝘅 𝗹𝗼𝘄𝗲𝗿 𝗰𝗼𝘀𝘁 𝘁𝗵𝗮𝗻 𝗖𝗹𝗮𝘂𝗱𝗲 𝗙𝗮𝗯𝗹𝗲 𝟱. We built FrontierFinance with a simple goal: improve AI for investment decision-making. After releasing the benchmark, our team identified opportunities to improve tool design, optimize context management, and increase reasoning effort. These changes improve performance across Screening & Discovery, Financial Data Extraction, Earnings & Events, and Company Research, helping Samaya extract quantitative and qualitative data, along with improved contextual reasoning and analysis. Full announcement below 👇