Inference.Daily
LLMs

Open Weights, Closed Margins: Reading Phi-4

Inside the quiet rewiring of research synthesis at HubSpot.

By Yuki Tanabe3 min read

If you spend enough time watching the AI industry, you stop reacting to launches and start tracking patterns. Open Weights, Closed Margins is one of those patterns.

Teams that win with tool-first agents tend to share a habit: they write the evals before they write the prompts. Everything else follows from that.

Inside Shopify, the rollout looked less like a moonshot and more like a slow migration. A pilot, a champion, a quiet expansion, a budget line.

The skeptical read is that we are watching a feature, not a platform. The optimistic read is that vision input is exactly the kind of feature that becomes a platform when nobody is paying attention.

The cost curve matters here. Mistral Large 3 is roughly an order of magnitude cheaper per token than the equivalent model 18 months ago, and that changes which problems are worth automating at all.

The skeptical read is that we are watching a feature, not a platform. The optimistic read is that voice mode is exactly the kind of feature that becomes a platform when nobody is paying attention.

Klarna has been quietly running QBR prep through Hex Magic for months. The results are unglamorous and, for that reason, more interesting than another benchmark chart.

None of this guarantees a clean story. Anthropic could ship a model next month that rearranges the assumptions in this piece. But the direction of travel, for now, is clear enough to plan around.

#evals#safety#multimodal#agents#vision

Related reading

More in LLMs

LLMs·

The Quiet Architecture Shift Behind Llama 4

Beyond the launch posts, DeepSeek V4 is reshaping how product managers approach contract review. We talked to the people actually using it in production.

Elena Brost3 min read

More from Inference Daily

Keep reading