OpenAI Launches GPT-6.1 Terra, Half-Price API with Improved Performance

Key Takeaways

1OpenAI halves the API pricing of GPT-6.1 Terra compared to GPT-5.6 Terra

2The model shows performance gains in software automation and coding

3Defenses reach 99.99% against direct attacks, with fewer hallucinations

Why it matters : GPT-6.1 Terra combines lower costs, increased efficiency, and enhanced security, broadening access to demanding professional use cases.

OpenAI is commercializing GPT-6.1 Terra with API pricing halved compared to GPT-5.6 Terra and expanded availability for ChatGPT Work and Codex offerings. Supported by benchmarks and security measures, the company and analysts report gains in software agents, coding, and hallucination reduction, while steering heavier usage towards higher models.

Prices Halved and Caching Options at $0.05 per Million

OpenAI announces that GPT-6.1 Terra's API prices are cut in half compared to GPT-5.6 Terra. Input tokens are charged at $1 per million, and output tokens at $5 per million. In comparison, Claude Sonnet 5.5 charges $2 for input and $10 for output, while GPT-5.6 Terra was priced at $2 and $12. Cached inputs drop to $0.05 per million tokens, a quarter of the previous rate. OpenAI positions Terra as a direct competitor to Anthropic's Sonnet series at half the price. The model is accessible to Plus, Pro, Enterprise, and Edu customers in ChatGPT Work and Codex, as well as via the API.

Software Automation: Cost per Task Up to Nine Times Lower

OpenAI highlights GPT-6.1 Terra for automated software control. On Agents' Last Exam, Terra matches the performance of GPT-6 Astra at one-tenth the token price. Against Claude Opus 5.5 at maximum effort, OpenAI reports more than tripled token efficiency and a quarter of the token price, resulting in a cost per task approximately nine times lower. The company adds that Terra completes tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol. In visual reasoning on MMMU-Pro, Terra outperforms GPT-5.6 Terra at all levels and approaches the results of GPT-6 Astra and GPT-6.1 Sol. On OSWorld 2.1, it surpasses the best score of GPT-5.6 Terra at medium effort for a fraction of the cost.

Le brief IA que lisent les pros

Le brief IA que les pros lisent chaque soir

Les 7 actus IA du jour, décryptées en 5 min. Gratuit.

Inclus dès l'inscription : notre sélection des meilleurs guides & comparatifs IA.

Choisissez votre rythme

Gratuit · Pas de spam · Désabonnement en 1 clic

Coding: 21.5% to 50.3% of Tasks Solved on Terminal-Bench 4.0

On Terminal-Bench 4.0, GPT-6.1 Terra solves 50.3% of tasks at maximum effort, compared to 21.5% for GPT-5.6 Terra. Artificial Analysis estimates that this gain of about 20 points places Terra at the level of GPT-6.1 Sol on this benchmark. On Terminal-Bench-Science 0.1, Terra outperforms GPT-6.1 Sol at a lower cost and is just five points behind Claude Opus 5.5 for less than one-fifth of the estimated cost. However, OpenAI recommends prioritizing GPT-6.1 Sol and GPT-6 Astra for the most demanding coding tasks.

Enhanced Security and 99.99% Defenses Against Direct Attacks

GPT-6.1 Terra achieves the "Critical" cybersecurity threshold defined by OpenAI's Preparedness Framework and inherits the protections of GPT-6 Astra. Compared to its predecessor, the model is described as more likely to adhere to its assigned limits and to clarify its restrictions to users. According to the system map, GPT-6.1 Terra, GPT-6 Astra, and GPT-6.1 Sol achieve a 99.99% success rate against direct attacks on the instruction hierarchy. For indirect prompt injections, GPT-6.1 Terra shows a 98.8% defense rate, up from 96.2% recorded for GPT-5.6 Terra.

Performance Metrics: 48 on the Intelligence Index and Fewer Hallucinations

Assessments indicate that GPT-6.1 Terra offers a better cost-performance ratio and a decrease in hallucinations. OpenAI claims to push the boundaries of cost-performance. Artificial Analysis refers to a new standard of intelligence per dollar: at maximum effort, Terra gains 6 points over GPT-5.6 Terra for a cost per task of $0.40 compared to $1.40, and at lower efforts, the gains range from 7 to 13 points, placing Terra on the Pareto frontier. The Intelligence Index credits Terra with a score of 48, between GPT-6.1 Sol at 52 and GPT-5.6 Terra at 42, and the cost-performance curves indicate it has the best intelligence/cost ratio at low effort. In the Coding Agents Index v2.0, Terra scores 48, six points below GPT-6.1 Sol and 17 above GPT-6 Luna. On the private benchmark AA-Omniscience, the model only "guesses" 38% of the time when it does not have the answer, compared to 88% for GPT-5.6 Terra, with an accuracy increase of two points. These improvements target office tasks and automated software control.