There’s no shortage of model launches these days, but this one stands out. OpenAI’s latest release, GPT‑4.1, promises something the legal field has been quietly anticipating: real gains in instruction following.

Legal prompts rarely stop at a simple “rewrite this clause.” They typically come embedded with commercial context and layered legal nuance. A straightforward request might sound more like: “Revise this clause to reflect a more favorable position for our client, but ensure consistency with the termination provisions and applicable law.” GPT‑4.1 shows a marked improvement in handling that kind of precision. For our upcoming Sonar Legal Word add-in, this means we can begin crafting tighter, more targeted prompts that actually do what you ask—without needing a dozen follow-ups.

Another critical advancement lies in GPT‑4.1’s handling of constraints, a known weakness in many models. When instructed to “Amend this clause, but don’t alter the liability cap,” the model is significantly more reliable at honoring that limitation. Importantly, when unsure, GPT‑4.1 is now more likely to acknowledge uncertainty—offering a clear “I don’t know” instead of fabricating a confident, incorrect answer.

On speed, GPT‑4.1 delivers modest gains over GPT‑4o. That said, in most legal applications, shaving off milliseconds matters far less than maintaining precision and fidelity to instructions. Accuracy remains the benchmark that matters most.

It’s worth noting that GPT‑4.1 is not available in ChatGPT—it’s an API-only release, built for integration into custom tools. That’s where we come in. We’re actively benchmarking GPT‑4.1 against other leading foundation models and will integrate it into Sonar Legal wherever it provides measurable advantages. Our goal remains constant: to ensure you always have access to the most capable model for the task at hand.

In Part 2, we’ll examine the section of the GPT‑4.1 technical paper focused on context retrieval from large documents. We’ll unpack how it handles the “needle in a haystack” problem, and what that means for high-stakes contract review.