OpenAI Unveils GPT‑6.1 Terra, Cutting Agent API Costs in Half

OpenAI released GPT‑6.1 Terra, a new multimodal model that combines fast computer‑use capabilities with vision, and priced the API at $1 per million input tokens for text‑only requests and $5 per million for image‑augmented queries. The move slashes the cost of running medium‑effort agents to about $0.29 per task, compared with the $4.82 price tag of the earlier GPT‑6.1 Sol model, according to MarkTechPost.

---

What is GPT‑6.1 Terra?

GPT‑6.1 Terra is the latest offering in OpenAI’s GPT‑6.1 family. It is described as a “fast computer‑use and vision model,” meaning it can process textual prompts, execute code‑like operations, and interpret visual inputs within the same request. The model is positioned as a versatile workhorse for applications that need both reasoning and image understanding, such as document analysis, UI automation, or visual‑search assistants.

OpenAI’s announcement highlights two pricing tiers:

• $1 per million tokens for pure text interactions.
• $5 per million tokens for requests that include image data.

These rates are significantly lower than the $2‑$10 range that OpenAI previously charged for comparable multimodal models, making Terra attractive for high‑volume workloads.

---

How does the new pricing compare to the previous GPT‑6.1 Sol?

The primary headline is the reduction of API costs for agent‑type tasks. MarkTechPost reports that a medium‑effort agent built on GPT‑6.1 Sol cost roughly $4.82 per task when evaluated on the standard "Last Exam" benchmark. By switching to Terra, the same benchmark score can be achieved for $0.29 per task – an almost 94 % price reduction.

The price cut stems from two factors:

1. Lower per‑token rates – Terra’s $1/$5 pricing is roughly half of Sol’s $2/$10 rates.
2. Optimized compute pathways – OpenAI claims Terra uses a more efficient inference engine, reducing the amount of GPU time required per token.

For developers who run large numbers of autonomous agents (e.g., customer‑support bots, workflow automators, or research assistants), the savings translate into dramatically lower operating expenses.

---

What does the performance look like?

OpenAI states that Terra matches the performance of GPT‑6 Astra on the “medium‑effort Agents’ Last Exam” benchmark. While the source does not provide a full accuracy table, the implication is that Terra delivers comparable reasoning quality while being cheaper and faster.

Key performance attributes mentioned include:

• Vision integration – Agents can ingest screenshots, PDFs, or camera feeds without a separate OCR step.
• Computer‑use speed – The model can execute code snippets or API calls within the same request, reducing round‑trip latency.
• Consistent scoring – Benchmarks indicate that Terra does not sacrifice reliability despite the lower price.

---

Who will benefit most from Terra?

1. Developers building autonomous agents – Lower per‑task costs make it feasible to run large fleets of bots that handle routine queries, data extraction, or workflow automation.
2. Enterprises with visual workloads – Companies that need to process invoices, product images, or UI screenshots can now do so at a fraction of the previous cost.
3. Researchers experimenting with multimodal prompting – The cheaper token rates encourage more extensive prompt engineering and iterative testing.
4. Start‑ups and indie creators – Budget‑constrained teams can now access high‑end vision‑enabled reasoning without blowing through credits.

---

How can developers start using GPT‑6.1 Terra?

OpenAI has added Terra to its standard API console. To begin:

1. Create or update an API key in the OpenAI dashboard.
2. Select the model name gpt-6.1-terra (or the exact identifier provided by OpenAI) in the request payload.
3. Adjust your token budgeting – Remember that image‑augmented calls count toward the $5 per‑million‑token tier.
4. Monitor usage – OpenAI’s usage dashboard now separates text and image token consumption, helping you stay within budget.

Developers can also experiment with the model through OpenAI’s Playground, which now offers a “Vision” toggle for uploading images alongside prompts.

---

Where does this fit into Grafy?

Grafy’s Bots app (see /bots) lets creators embed AI agents directly into websites, support portals, or internal tools. By selecting the model in the Bot configuration panel, users can now point the Bot to GPT‑6.1 Terra, gaining both vision capabilities and the reduced cost structure.

For developers who prefer a no‑code environment, the Chat app (/chat) also supports model switching, allowing quick side‑by‑side comparisons of Terra versus other OpenRouter or Vertex AI models. This makes it easy to prototype agent workflows, test visual prompts, and evaluate cost before committing to production.

If you are primarily interested in reading the technical details or benchmarking the model yourself, Grafy’s Reader (/reader) can ingest the MarkTechPost article and other PDFs, letting you ask natural‑language questions about the model’s specs without leaving the platform.

---

What are the broader implications for the AI market?

OpenAI’s price cut signals a competitive response to other providers that have recently lowered their own LLM rates (see the earlier “OpenAI and Anthropic cut LLM prices by 80%” coverage). By making high‑performance multimodal agents cheaper, OpenAI may accelerate adoption in sectors that were previously hesitant due to cost, such as legal tech, medical imaging triage, and large‑scale knowledge‑base construction.

The move also pressures rivals to improve efficiency. If Terra’s cheaper compute pipeline proves scalable, we could see a wave of model‑optimisation efforts across the industry, potentially leading to even lower prices and greener AI workloads.

---

Frequently asked questions

How does the $5 per‑million‑token image rate compare to other vision models?

OpenAI’s $5 rate is roughly half of the $10‑$12 range that many competing vision‑enabled LLMs charge. It places Terra among the most cost‑effective options for mixed text‑image workloads.

Can existing GPT‑6.1 Sol applications switch to Terra without code changes?

Yes, in most cases the only change required is updating the model identifier in the API request. However, developers should audit token usage because image‑enabled calls now fall under a separate pricing tier.

Is there a usage cap or quota for Terra during the early rollout?

OpenAI has not announced specific caps, but new models often start with a soft launch that includes rate limits. Users should monitor the dashboard for any quota alerts and contact OpenAI support if they anticipate high‑volume traffic.

---

Source: MarkTechPost — OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work