No idea who's actually worth more than who here, but I will say that recently I've stopped using Opus at work for the lion's share of technical work. I used GPT-6.1 Sol for one big project I tackled last week, where the agents spent about 20 hours to produce ~73k LoC across 22 PRs. While the technical decisions and code weren't bad and design gaps were left clearly enumerated, the overall "human factor" was terrible, in the sense that the plan docs were overly-verbose and difficult to read and PR descriptions were abysmal. Numbers and words were run together all over the place where there should have been spaces. At that point I pointed Opus 5.5 at the whole thing and asked it to double-check design and code and rewrite all the "for human consumption" text in plan docs and PR description, and it did a phenomenal job of that part of it. Plan docs were made much more concise with emphasis on the strategic parts of the problem. Opus accepted the vast majority of the design and code that GPT-6.1 Sol had produced.

Of course I have no idea what the technical results would have been if I had just used Opus 5.5 end-to-end, but I'm fairly confident it would have been more expensive given the rates my company pays for each. For that matter I wonder how Kimi3.1 Pro or DeepSeek 4.1 Flash 1017 would have done with various parts of the job. In fact I may try rerunning the whole thing using open weight models next week just out of curiosity.

If Kimi+DeepSeek produce something roughly equivalent, then just from my own personal experience with a novel problem involving legacy in-house code, as opposed to public benchmarks for which I don't have a great handle on how well they correlate with value produced by the models, that will make me question the valuation of both OpenAI and Anthropic.

I use Opus 5.5 for planning, sprinkle in a Fable 5.1 review of the spec (they go back and fourth), then send the spec and plan to my local deepseek 4.1 Flash 1017 to implement. It crunches away for a good while (using ds harness)

Then check w/ Opus and Astra for good measure if it really matters. Works well. DSv4.1 is okay but does make mistakes. I run my 2x DGX sparks at 60-70% capacity this way, which nets about $400/mo worth of tokens on $30 of electricity.

I am hoping in the next six months to be able to in-house ALL of the above using a slightly more competent model, or two more sparks. It is slow running local, but it is really fascinating to have all this work done from my desk.