OpenAI shipped GPT-6.1 Terra for computer use: 15 points above GPT-5.6 Terra on Agents' Last Exam at one-third the cost, and over three times faster than GPT-6 Astra
AspireAI community · OpenAI
OpenAI introduced GPT-6.1 Terra on October 26, a model built for visual reasoning and computer use - the kind of workday work where an agent drafts proposals, interprets technical drawings, fills in forms or edits video. Standard API prices are 50% lower than both GPT-5.6 Terra and GPT-6.1 Sol, and it ships as gpt-6.1-terra in the API plus Plus, Pro, Enterprise and Edu plans in ChatGPT Work and Codex.
On Agents' Last Exam, at medium reasoning effort Terra scores 15 percentage points higher than GPT-5.6 Terra at about one-third the cost, and performs on par with GPT-6 Astra at one-tenth its price per token. At max effort, OpenAI reports it is over three times as token efficient as Claude Opus 5.5 at one-fourth the price per token, roughly 9x lower cost per task, and completes tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol. On Terminal-Bench 4.0 at max reasoning it solves 50.3% of tasks, more than double GPT-5.6 Terra's published 21.5%. On OpenAI's internal error-inducing-prompt benchmark, the error rate at low reasoning effort falls from 53.6% to 16.7%, and at max effort by 18 points to 4.3%.
The caveat is that every number above is OpenAI's own, the error-rate benchmark is internal and not public, and OpenAI itself says GPT-6.1 Sol and GPT-6 Astra remain the better choice for the most complex coding tasks. Pricing is $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens.
https://openai.com/index/introducing-gpt-6-1-terra
Read the story and explore more from the AspireAI community.
Two sparse-autoencoder analyses have resisted scaling: matching semantically similar features across layers, and compressing large feature circuits into interpretable supernodes. This paper argues both are the same underlying problem, estimating semantic distances between SAE features that sit on different activation manifolds. Its representation treats each feature not as a single decoder vector but as an activation-weighted distribution over the hidden states expressing it, projects those distributions into a shared reference space, and compares them with Wasserstein distance. The representation is proved invariant to activation rescaling and stable under perturbation, and recovers true matches under finite-sample margin conditions. Empirically it outperformed decoder-vector and LLM-based baselines and captured subtle functional distinctions between related features, and it compressed large feature circuits into interpretable supernodes automatically. The limitation is that this is interpretability tooling: the paper demonstrates better feature matching and circuit compression, not that downstream interpretability claims become more accurate.