Anthropic and OpenAI launch cost-efficient agent models

The shift to agentic workhorses

The conversation around AI models has shifted from simple chatbot interactions to the economics of autonomous agents. As of October 24, 2026, the primary concern for developers is which model finishes the most real work for the lowest price. Anthropic and OpenAI have both responded by launching models specifically designed for these high-volume, multi-step tasks.

Anthropic introduced Claude Opus 5.5 as its flagship model for the new 5.5 family. Simultaneously, OpenAI released GPT-6 Sol and GPT-6 Luna into their API and ChatGPT Work platforms. These models are positioned as the primary tools for professional work, including coding and complex computer use.

Both companies are leading with aggressive pricing to capture the mid-tier market. While premium models like GPT-6 Astra remain available for high-end needs, Sol and Opus are intended to be the models that most teams run all day. This move signals a transition toward making AI agents commercially viable for large-scale deployment.

Comparing the price ladders

Each lab now offers a three-step ladder of models, but the rungs do not align perfectly. Anthropic's ladder includes Opus 5.5 as the flagship, Sonnet 5.5 as the mid-tier, and Haiku 5.5 as the high-volume small model. OpenAI's ladder features GPT-6 Astra at the top, GPT-6 Sol in the middle, and GPT-6 Luna as the small, cheap tier.

The pricing relationship between the two main workhorses is remarkably symmetrical. The list price for Opus 5.5 is double that of GPT-6 Sol across input, output, and cached tokens. This 2-to-1 ratio simplifies the math for developers who need to compare the two across different types of usage.

Because the sticker prices are so consistent, the debate moves away from the price per token and toward the number of tokens required to finish a job. If Opus 5.5 is more efficient and uses fewer tokens per task, it could potentially be cheaper than Sol despite the higher list price. This makes the performance of the model's logic a key factor in the final bill.

Navigating the data gaps

A significant hurdle for developers is that no primary source has published a clean, head-to-head test of Opus 5.5 against GPT-6 Sol under identical conditions. Anthropic chooses to benchmark its new model against the more expensive GPT-6 Astra, while OpenAI pits its Sol model against the older Claude Opus 5. This means both labs are selecting the rival that makes their own specific marketing claims look the strongest.

Because of these mismatched comparators, both companies' statements can be true at once without actually answering how the two current workhorses compare to each other. To find clarity, third parties are beginning to crowd into the space to fill these gaps. For example, the Kimi K3 coding leaderboard shows how quickly an open-weight challenger can scramble a ranking that previously looked settled.

Stable pricing and new updates

For those tracking OpenAI's specific releases, the price points for GPT-6 Sol remain stable regardless of the version. The $2 and $10 price points are the same for both GPT-6 Sol and GPT-6.1 Sol, which ensures the dollar comparison remains steady for users who might switch between these specific revisions. This stability helps teams plan their budgets even as the labs continue to roll out different iterations of their middle-tier models.

To stay updated on these shifting benchmarks and new tech stories, readers can add Shattered once in the Google app. This ensures that new stories appear in news suggestions, providing a way to track how these model rankings and price points evolve in real time as new checkpoints land.