Anthropic Releases Claude Sonnet 5 — Near-Opus Performance at Sonnet Pricing
On June 30, 2026, Anthropic announced its new model, Claude Sonnet 5. It outperforms the previous-generation Sonnet 4.6 across coding, agentic tasks, and professional work, and on some benchmarks it posts scores that surpass even the higher-tier Opus 4.8. Specifically, it delivered a score of 1,618 on the GDPval-AA v2 test measuring knowledge work, edging out Opus 4.8 by a narrow margin.
Its biggest highlight is strengthened agentic capability. The model can autonomously operate tools such as browsers and terminals, handling at Sonnet pricing tasks that just a few months ago required larger, more expensive models. On the pricing front, it is set cheaper than not only Opus but also GPT-5.5 and Gemini Pro, positioning it to help cut the costs of running agents. It also became available the same day on Amazon Bedrock and the Claude Platform on AWS.
Sonnet-class models have historically led the era of agentic AI (Sonnet 3.5, 3.6, and 3.7 were the first to reach a practical level in coding and tool use), and this Sonnet 5 can be seen as the latest milestone in that lineage.
Research Workbench Claude Science Released the Same Day
Alongside Sonnet 5, Anthropic also launched Claude Science, an AI workbench for scientific researchers, on the same day. Its defining characteristic is a focus on workflows rather than a new model. It is designed to reduce the burden on researchers who shuttle between databases, pipelines, and various tools, letting them complete computational research within a single environment.
It ships with more than 60 preconfigured skills covering fields such as genomics and computational chemistry, and it also includes agents that automatically verify paper citations and computational results. It runs not only in local environments but also on HPC clusters. It can automate biology and chemistry tasks such as protein structure prediction, and it is expected to trim the boring parts of research.
For now, Claude Science is offered as a customizable app, with a mechanism for integrating the tools and packages researchers use daily.
OpenAI Announces GeneBench-Pro — An AI Benchmark for Genomics and Biology
On the same day, OpenAI announced GeneBench-Pro, a new benchmark for evaluating AI performance in genomic analysis, biology, and scientific research. Its distinguishing feature is the use of complex real-world datasets to test how useful AI models actually are in real scientific research settings. Case studies have also been published, demonstrating the benchmark's value through concrete application examples.
The development of evaluation standards for AI applications in science is significant when it comes to selecting models and incorporating them into research processes. The succession of science-focused tools arriving in step with Anthropic's Claude Science suggests that the AI industry as a whole is starting to pour effort into scientific research.
Google Adds Nano Banana 2 Lite and Gemini Omni Flash to Its API
Google has added two new generative AI models to its API. Nano Banana 2 Lite is an image generation model that produces images in 4 seconds at a low cost of $0.034 per image. The other, Gemini Omni Flash, makes video generation and editing from text prompts available via API for the first time.
Google recommends chaining the two models, enabling workflows such as generating an image from a simple text input and then turning it into a video. Faster and cheaper image generation is an ongoing trend, and it is likely to influence creators' production processes.
OpenAI Halves Inference Costs for Guest ChatGPT — The Information Reports
According to a report by The Information, OpenAI has cut its AI model inference costs by more than half. After the optimizations were applied to ChatGPT, the number of NVIDIA GPUs required reportedly dropped temporarily to the hundreds. With the cost of running AI models still a major challenge, this move shows there is still substantial room for efficiency gains.
The cost reduction is believed to be part of its handling of guest users (those who use the service without logging in), suggesting that OpenAI is improving infrastructure efficiency to maintain and expand its free tier.
---