OpenAI released GPT-6.1 Terra on October 26 at half the per-token price of GPT-6.1 Sol, and it sits in an awkward position. What GPT lacks against Claude is a frontier model, not an affordable mid-range one.

On Artificial Analysis’s Intelligence Index, GPT-6 Astra scores 57 at max, higher than Sol’s 55 and the rest of GPT. But at $6.24 per task against Sol’s $1.27, its price makes it a “specialized model”, just like Fable 5.1 on Claude. Both list at $10 per million input tokens and $50 per million output. We can’t equate Sol to Opus 5.5, but in their own lineups, they take up similar spaces.

The top is where GPT falls short. Opus 5.5 leads the Index at 58, a point above Astra, and Opus is not Claude’s frontier model. Fable 5.5 is not even out yet.

Further down, both lineups are covered. Claude now has Sonnet 5.5 and Haiku 5.5 to fill in cheaper models, and on GPT, the new Terra and Luna seem to fare fine. At medium to high, Luna is a cheap model with a definite ceiling, scoring 18 to 21 for $0.04 to $0.07, while Haiku scores 25 to 30 for $0.09 to $0.16. Overall, the zone GPT’s cheaper models cover is relatively lower than the one Claude’s cover. They are competitively priced at similar scores, but is it worth subscribing to both just for a subagent?

The few use cases I’ve seen are as narrow as Fable’s, such as using Terra for computer use. OpenAI pitches it as a model that combines “detailed visual reasoning and precise computer use”. There, it would be cheaper than Sol and still more capable than Luna. On the Intelligence Index scale, it is cheaper than Sonnet 5.5 at similar scores. I haven’t incorporated it into my workflow yet, and I’m not entirely sure spinning up a subagent just for computer use saves tokens either. It could be too niche, or it could be just the right tool vibe coders wanted.

Both Claude and GPT need Fable 5.5 and Astra 6.1. The new Astra was planned for October and delayed in late September, and whenever it lands, it has Opus 5.5 to compete against, not to mention a potential Fable that will score higher than the standing champion. For a frontier model, many would pay the price for more capability. GPT still has a positioning problem, and a cheaper mid-range model is not what it was missing.