Optiver

Anthropic has released Claude Opus 4.8, and our headline finding is clear: performance continues to improve, especially at lower reasoning-effort settings. On Optiver's trading intern exam, Opus 4.8 delivers its strongest gains where less reasoning compute is required. That's an encouraging sign that frontier models are becoming both more capable and more efficient. The charts below show two views of that progress. The first tracks performance across twelve Claude model releases, from Claude 3 Haiku to Opus 4.8. The second compares Claude Opus 4.6, 4.7 and 4.8 across different reasoning-effort tiers. Read more about our research into AI and trading: https://lnkd.in/efqm-uPq

Claude Opus 4.8 is out, and we've been testing it on some of our real world trading problems as we introduced here: https://bit.ly/4uINgjw This chart shows how Opus 4.8 scores on the trading internship exam we use at Optiver, across different reasoning-effort settings and relative to previous generations of Claude models. It's exciting to see the continued progress on this exam, particularly at lower effort settings. Congrats to the Anthropic team on the release.

Big congrats to the Anthropic team on the Claude Opus 4.8 launch. We’ve been evaluating it ahead of today’s release and have seen promising results in our evals so far. Looking forward to what engineers build with it.

Teams building and scaling better agentic outcomes are building systems designed for agents to check their own work, and fix it, instead of relying on humans to detail corrections. At AI Engineer Melbourne next week, Moss Ebeling will speak about what closed-loop design looks like and how you can use it to build real leverage via automation in your workflows. If you’re attending, check out this AI Engineering track event on Wednesday at 1:30pm and stop by the Optiver booth to talk with the team.

We recently shared our vision for technology at Optiver: what we’re building, where we’re headed, and how it all fits together. Platform Engineering is a core part of that. The infrastructure, developer experience, data, and AI layer all shape how quickly teams can move. I’ve been thinking a lot about how agentic AI changes not just how we build, but what we build for. I recently sat down to talk through that and what Platform Engineering means at Optiver now. Worth a read if you’re interested in the space: https://bit.ly/4tObEim

“We have a tremendous advantage at Optiver in that we own our path to production.” That line from our most recent Tech Town Hall reflects how we think about engineering: direct ownership, tight feedback loops, and work that connects quickly to live outcomes. As we scale into more markets, products, and systematic strategies, the focus is clear: build reusable platforms, scale agentic AI workflows with strong guardrails, and help teams move from idea to impact faster. Learn more about tech at Optiver: https://bit.ly/4dhmdph

It was a pleasure speaking at the BMLL conference yesterday about a fundamental shift we’re seeing in our industry: the transition from systematic automation to agentic systems. In today’s markets, the firms that win are those that can learn, iterate, and operationalize insight faster than the competition. Here are  four key takeaways on where we are headed: • The Agentic Shift: We are moving beyond deterministic, rule-based systems. We’re entering an era where systems can reason, adapt, and synthesize information autonomously. This doesn't replace humans; it elevates our role in governance and strategy. • Data is the Bedrock: Agentic systems are only as good as their data that fuels them. As decision-making speeds up, the demand for high-fidelity data is becoming exponential. You cannot have autonomous intelligence without trusted context. • Build vs. Partner: Differentiated engineering talent should not be spent on commodity problems. Strategic partnerships (like ours with BMLL) allow our researchers to stop worrying about data normalization and start focusing on higher-order alpha generation. • Culture Over Code: Stability remains the "first feature." Whether you’re building cloud-native research environments or AI-assisted workflows, a culture of ownership and discipline is what actually scales. The next generation of competitive advantage will be won by the firms that succesfully combine high-quality data, scalable compute, and AI First workflows.  Thanks to the BMLL team for hosting a great event. Looking forward to seeing how this "Agentic Odyssey" continues to reshape global liquidity. #FinTech #AI #MarketMaking #Optiver #EngineeringLeadership #DataScience #CapitalMarkets

Interesting FPGA problems are rarely isolated to the FPGA. As trading systems scale, latency and performance constraints emerge across hardware, software, and protocols. Solving them means understanding the full system well enough to decide what should change and where. At IEEE International Symposium On Field-Programmable Custom Computing Machines (FCCM), Optiver FPGA Engineering Lead Kevin Sprague will discuss how our engineers approach system-level tradeoffs in low-latency trading infrastructure. If you can’t make the talk, sign up to receive the recap: https://bit.ly/4tqnMGc

Behind high-performing teams are strong connections. Our US teams came together in Quebec City for a weekend of skiing, tobogganing, recharging and exploring the city. It was a chance to step away from the desk, connect across offices, and strengthen the relationships that carry into how we work every day. See the highlights from the trip.