OpenAI cuts agent model prices with GPT-6.1 Terra as Ant launches Lingguang 2.0

OpenAI said GPT-6.1 Terra costs $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. That is half the price of GPT-5.6 Terra and GPT-6.1 Sol and one-tenth of the flagship GPT-6 Astra's token price, the company said. The model is available to ChatGPT Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API as gpt-6.1-terra.

In the Agents' Last Exam test for professional agent work, GPT-6.1 Terra at medium reasoning scored 15 percentage points higher than GPT-5.6 Terra while costing about one-third as much per task, according to OpenAI. OpenAI said the model matched GPT-6 Astra on that test at one-tenth the token price. It also said the new model completes tasks three times faster than GPT-6 Astra and GPT-6.1 Sol. A developer test cited by Ifanr showed generation at 450 tokens per second, with an unverified claim of up to 600 tokens per second at low reasoning.

OpenAI said GPT-6.1 Terra reduced factual errors in its internal high-difficulty test to 16.7% at low reasoning from 53.6% for GPT-5.6 Terra, and to 4.3% at maximum reasoning. In Terminal-Bench 4.0, which tests command-line coding by agents, GPT-6.1 Terra reached a 50.3% task resolution rate at maximum reasoning, more than double the 21.5% public score for GPT-5.6 Terra. On Terminal-Bench-Science 0.1, it beat GPT-6.1 Sol and trailed Claude Opus 5.5 by fewer than 5 percentage points at less than one-fifth the estimated cost, according to OpenAI.

Artificial Analysis' Intelligence Index gave GPT-6.1 Terra (Max) 48 points, above Chinese open-weight models other than Kimi K3.1 Pro, which scored 55, according to Ifanr. OpenAI said the model exceeded GPT-5.6 Terra on MMMU-Pro visual reasoning at all reasoning levels and, on the OSWorld 2.1 desktop-operation test, beat GPT-5.6 Terra's maximum-reasoning score at medium reasoning. OpenAI said GPT-6.1 Terra reached the Critical risk level for cybersecurity under its Preparedness Framework and uses the same safeguards as GPT-6 Astra, with GPT-6-series alignment improvements for authorized scope, capability limits and prompt-injection resistance.

Ant Group's Lingguang 2.0, also released Oct. 27, is designed to move beyond single commands by identifying needs, tracking complex matters and reminding users at key points, according to Ifanr's review. If a user says she will travel to Shanghai for three days next week, the agent books flights and hotels, arranges the schedule, reminds her to leave for the airport and bring ID, and readjusts hotel check-in and meeting plans after a flight delay, the review said.

In a pet insurance test, Lingguang read a policy, listed reimbursement scope and required documents, suggested when symptoms warranted a hospital visit and set a follow-up reminder, retaining earlier details in later conversations, according to Ifanr. The agent also filtered DingTalk group messages for items needing a response, tracked Double 11 prices after receiving screenshots, recalculated rankings under a 500-yuan budget, compared platforms and sent a price-change alert the next day. Ifanr said tasks continued on a cloud host with a browser after the app was closed, and that Lingguang scanned Alipay bills for recurring charges, finding three cloud-storage subscriptions, two of them unused.

In a weekend trip plan for two adults and a 4-year-old, with a two-hour drive limit, a child nap, no spicy food, a Sunday 8 p.m. return and a 400-yuan per-person budget, Lingguang produced an itinerary and asked key questions, according to the review. When one adult canceled, it kept the framework, removed arrangements for the child and second adult, and added a new destination for a solo traveler, the review said.

Lingguang 2.0 connects to Alipay for bills, credit cards, Huabei, Jiebei, travel and other services, as well as QQ Mail, calendars, alarms, SMS, files, DingTalk, Feishu, Didi, Baidu Netdisk, Mijia, 12306 and car systems from Li Auto, Nio, Xpeng and Huawei, according to Ifanr. The review described Meta's Muse, launched Sept. 8 and downloaded more than 10 million times in 40 days, as an 'errand-type' agent for email and purchases, while positioning Lingguang as a 'management-type' agent for daily chores. Ifanr said Muse has integrated Stripe's Link payment and received its shopping guarantee, while Lingguang's tie to Alipay gives it access to payment and life-service scenarios.