Claude vs. OpenAI's GPT 6.1 Sol

TLDR: 6.1 Sol on xhigh effort is the best value available right now, but it is much slower than Opus 5.5 on high effort. Opus 5.5 high effort costs about 4.7× more for a bit smarter results, but at roughly a third the wait time. Looking at the rankings from Artificial Analysis, it is interesting that OpenAI is really challenging Anthropic on value. The data, however, is a bit more complex than what AA's charts show. Opus 5.5 and Fable 5.1 are the smartest models and also the most expensive per task. That's no surprise to me, so I wasn't very interested in the new GPT-6.1 Sol. I value quality. What caught my eye, however, is Sol's value. To judge that, however, you can't rely on the AA cost graph, below. That is because you shouldn't run either model at max effort. The sweet spot is high effort for Opus 5.5 and xhigh for 6.1 Sol. https://preview.redd.it/fgub9sdp0wsh1.png?width=980&format=png&auto=webp&s=5166969238219c0a789d5b13dac0b9c71267eb26 Analysis via Opus 5.5 Opus 5.5 (High) vs GPT-6.1 Sol (Xhigh): Artificial Analysis numbers Metric Opus 5.5 (high) GPT-6.1 Sol (xhigh) Intelligence Index 54 51 Cost per task $1.82 $0.39 Output speed ~73 t/s ~64 t/s Time to first token ~39s ~108s Output tokens (full index run) 53M 36M List price (in/out per 1M) $4 / $20 $2 / $10 Cost: Sol xhigh costs about 21% of what Opus high does, for 3 fewer index points. That's about $0.48 per extra point if you go with Opus. At 10,000 tasks a month, the gap is roughly $14,000. Sol is also the more concise model, using about a third fewer output tokens across the same eval run. Latency: Opus still wins clearly, though by less than it did against Sol max. Opus high reaches its first token in about 39s, compared with about 108s for Sol xhigh, so it's roughly 2.8× quicker. Sol max was around 273s, which means dropping to xhigh cuts Sol's wait by more than half while costing only 1 index point. Coding: In Artificial Analysis's Coding Agent Index snapshot, Codex with GPT-6.1 Sol (xhigh) scores 62.9%, compared with 66.0% for Claude Code with Opus 5.5 (max). I couldn't find a published Coding Agent score for Opus 5.5 at high effort, so the coding-specific matchup here isn't fully apples to apples. Matching on intelligence instead: Opus 5.5 on medium also scores 51 on the index, the same as Sol xhigh, but costs $1.34 per task, about 3.4× as much. At equal intelligence, Sol is clearly cheaper. Bottom line: If you're running high-volume coding agents where cost per task matters most,* Sol xhigh is probably the best value available right now*. Opus 5.5 high costs about 4.7× more but gives you 3 more index points and roughly a third of the wait time, which is worth it for harder tasks or when someone is waiting on the result. Source: Artificial Analysis