OpenAI releases GPT-6.1 Terra with stronger computer agent capabilities at lower cost than Claude
Article excerpt
OpenAI's new GPT-6.1 Terra model improves computer-use agents with better reasoning and tool handling while undercutting Anthropic's Claude on API pricing for agent workloads. OpenAI shipped GPT-6.1 Terra on Monday, a model update focused on computer-use agents that can navigate operating systems, manipulate files, and execute multi-step workflows with less supervision. The company claims the model matches or exceeds Anthropic's Claude 4 Opus on agent benchmarks while cutting API costs by roughly 40 percent for typical automation workloads. The release targets developers building autonomous agents that operate desktop environments, web browsers, and command-line interfaces. GPT-6.1 Terra introduces a revised function-calling schema that reduces hallucinated tool calls by 62 percent compared to GPT-6, according to OpenAI's technical report. The model also adds native support for structured output validation, letting developers define JSON schemas that the model must satisfy before returning tool results. "The big shift is reliability," said Maria Chen, a senior engineer at Browserbase who tested the preview. "Previous models would get stuck in loops or hallucinate file paths. Terra stays on task longer and recovers from errors without human intervention." Pricing for GPT-6.1 Terra starts at $8 per million input tokens and $24 per million output tokens. Anthropic's Claude 4...
Keep reading with a free account
The rest of this article, and every signal for OpenAI, is in your free account.
Extracted from this sentence
The model integrates with OpenAI's Agents SDK, updated today to version 0.4.
