Two $2/$10 Launches, One Clear Winner on the Index
Claude Sonnet 5.5 and GPT-6.1 Sol shipped a day apart at the same $2/$10. On the Artificial Analysis index Sonnet leads by four points and sits two under Opus 5.5 at half the price.
Anthropic and OpenAI shipped mid-tier models a day apart at exactly the same price: $2 per million input tokens and $10 per million output. Claude Sonnet 5.5 arrived on Sep 28, and GPT-6.1 Sol followed on Sep 29 at OpenAI DevDay. The price is the same, but the scores are not. On the Artificial Analysis Intelligence Index at max reasoning, as reported by The Batch, Sonnet 5.5 scores 56 and Sol scores 52.
Sol looks best next to its own family. It costs one-fifth of GPT-6 Astra's $10/$50, and The Batch reports that it lands within a point of Astra on the index (52 against 53) at less than a quarter of the cost per task. It is also four points ahead of GPT-6 Sol's 48. Treat the decimals loosely, though. Another blog puts Astra and Sol at 52.7 and 51.8, so a one-point gap tells you less than it seems to.
Sonnet 5.5 keeps Sonnet 5's price and sits two points below Claude Opus 5.5's 58. helloai tracks Opus 5.5 at $4/$20 with a 1M-token window, so Sonnet gets most of the flagship's score at half the list price. Anthropic reportedly claims up to 30% lower cost per task because Sonnet 5.5 uses fewer tokens. That claim and the pricing come from third-party aggregators, and we did not check Anthropic's own page. Both new models also charge less for output than Gemini 3.1 Pro, our other tracked reference point, at $10 against its $12.
Some caveats apply. helloai does not ingest the Artificial Analysis index. LMArena text-overall Elo remains our only ranking signal, and any index is just one view of a model. Neither mid-tier model is in our tracked set, which is capped at six with no model dropped. Sol does appear on the October 8 arena.ai text board, which we fetched on Oct 12: gpt-6.1-sol-max sits at 1484 (interval 1475 to 1493) on 4,512 votes, its second straight snapshot. That puts it one point above tracked Qwen3.8-Max at 1483, well behind Opus 5.5 at 1507 and well ahead of Grok 4.7 at 1443.
This is bad news for flagship models. When a $2/$10 model comes within two points of its $4/$20 sibling, or within one point of a $10/$50 one, the flagship is only worth paying for on tasks where those last few points make a measurable difference. Teams setting routing defaults this month should expect the mid-tier to take most of their traffic. The next round of flagship pricing will have to justify itself on a smaller slice of hard workloads.