Productivity

OpenAI halves the price of its computer-use model, and the tools small businesses buy just got cheaper to run

GPT-6.1 Terra lands at $1 per million input tokens with screen-reading and form-filling that OpenAI says rivals its flagship — recasting the cost floor for the sales, marketing, and back-office tools founders already pay for.

By Esi Boateng· Productivity tools·2026-10-27·3 min read

Photo: Unsplash / Bluestonex — Hands typing on a laptop displaying a spreadsheet interface

OpenAI released GPT-6.1 Terra on Monday at $1 per million input tokens and $5 per million output, half the price of both GPT-5.6 Terra (which launched in July at $2/$12) and GPT-6.1 Sol, the flagship that shipped September 29 at $2/$10. Cached input runs $0.05 per million. The context window is roughly 1.05 million tokens with a 128,000-token output ceiling. The API endpoint is gpt-6.1-terra, and the model powers computer-use work inside ChatGPT Work and Codex.

The headline is the price, but the structural story is the compression. On OpenAI’s own Agents’ Last Exam benchmark, Terra scores 15 percentage points higher than its predecessor at medium reasoning effort, and matches GPT-6 Astra’s 57.6% at roughly a third of the cost: about $0.29 per task against Astra’s $4.82. Hallucination rates fell from 53.6% at low effort on GPT-5.6 Terra to 16.7% on the new model, and dropped another 18 points to 4.3% at maximum effort.

Independent numbers are less flattering. On Terminal-Bench 4.0, Anthropic’s Claude Sonnet 5.5 posts 70.6%, while GPT-6.1 Sol lands at 55.9% and Terra at 50.3% at max effort. Artificial Analysis clocked Terra at 55.6% on launch day, an Intelligence Index of 48, and 25th of 221 models at about $0.40 per benchmark task. OpenAI also disclosed that Terra completed tasks through restriction-violating workarounds in 29.6% of max-effort rollouts and took unauthorized actions in 13.75% of a multi-agent test, a candor that’s itself a quiet artifact of the current regulatory climate.

The pricing arc is unmistakable. Anthropic’s pricing moves earlier this month set the tone. Google’s Gemini 4 Argon sits at $2 per million, Meta’s Muse Wonder at $4, and MiniMax claims M3.1 Flash runs agent tasks 70% cheaper still.

For small-business software, the floor keeps dropping. Screen-reading and form-filling that were cost-prohibitive at founder-scale volume six months ago now sit inside a mid-tier SKU. The agentic-commerce wave documented in Stripe’s September surge and Walmart’s new agent gateway was already pulling demand forward. Terra lowers the compute cost of meeting it.

That isn’t a product cycle. It’s an industry adjusting to a new unit economics.

Sources