Anthropic Launches Claude Opus 5.5, OpenAI Ships GPT-6 Sol and Luna at Half Price, and Claude Agents Find a CRISPR-Like Enzyme System

Posted on by

This brief covers the trailing ~72 hours (September 22–25, 2026). Every item below was confirmed on the originating organization’s own page, with a published date inside the window. Two frontier releases landed on the same day: Anthropic introduced Claude Opus 5.5, the first model in its Claude 5.5 family, and OpenAI extended its GPT-6 generation downmarket with GPT-6 Sol and GPT-6 Luna at 50% lower API prices. Anthropic also unveiled a life sciences lab and early results in which Claude agents found a previously unrecognized enzyme system. At the UN, Sam Altman addressed the Security Council and OpenAI extended its Daybreak cyber-defense program to Ukraine, and Google added real-time video avatars to Gemini 3.8 Live.

Anthropic introduces Claude Opus 5.5: Fable 5.1-level performance at 40% lower cost than Opus 5

Anthropic · September 22, 2026

Opus 5.5 is the first model in the Claude 5.5 family and Anthropic’s first release since it called for “pacing the frontier.” Anthropic reports 66.4% on Terminal-Bench 4.0, 54.4% on FrontierCode v1.1, 57.8% on CursorBench 4.0, and a GDPval-AA v2.1 Elo of 1846, ahead of Fable 5.1, Opus 5, and GPT-6 Astra on most of those measures. Pricing drops to $4/$20 per million input/output tokens with cache reads at $0.20 (60% below Opus 5), output is over 30% faster, and the model scored best to date on Anthropic’s automated behavioral audit. Because its biology and cyber capabilities are comparable to Claude Mythos 5.1, it ships with Fable 5.1-class safeguards that transparently fall back to other models; Sonnet 5.5 and Haiku 5.5 are due in the coming weeks.

“It performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5.” — Anthropic

Source: Introducing Claude Opus 5.5

OpenAI releases GPT-6 Sol and GPT-6 Luna, cutting API prices 50% versus GPT-5.6

OpenAI · September 22, 2026

OpenAI expanded the GPT-6 family with two cheaper tiers trained with methods similar to GPT-6 Astra. GPT-6 Sol is priced at $2/$10 per million input/output tokens and GPT-6 Luna at $0.10/$0.50, both half the GPT-5.6 promotional prices. OpenAI reports Sol scoring 33.2% on Zapier’s AutomationBench at $0.27 per task and 68.8% on DeepSWE v1.1, and says Sol makes about half as many factual mistakes as its predecessor on its internal evaluation. The launch also brings improved prompt caching with 90% discounts on cached reads and cache-preserving changes to reasoning effort and tools; the models are live in ChatGPT Work, Codex, and the API as gpt-6-sol and gpt-6-luna.

“GPT-6 Astra introduced a new generation of intelligence—these models help distribute the benefits of that intelligence by advancing the frontier on cost efficiency.” — OpenAI

Source: Introducing GPT-6 Sol and Luna

Claude agents discover a novel enzyme system with CRISPR-like repeats as Anthropic unveils a life sciences lab

Anthropic · September 23, 2026

Anthropic introduced a new life sciences research group and Bay Area wet lab focused on AI-driven genome mining. In its first reported result, roughly 950 Claude agents spent 21 hours and 210 million tokens surveying over 200,000 reverse transcriptases, narrowing 3,500 candidate systems to 20 detailed reports. One agent spotted a tandem repeat array next to an unusual RT in jumbo phages, a system Anthropic calls array-associated reverse transcriptases (ART). Initial lab experiments show the array is expressed as distinct short RNAs, suggesting a possibly programmable mechanism; its function is still under investigation, and a pre-print has been released.

“This is an exciting example of how AI agents can contribute to biological discovery.” — Feng Zhang, MIT and the Broad Institute

Source: Claude discovers a novel enzyme system with CRISPR-like repeats

Sam Altman addresses the UN Security Council, calling for international frontier AI standards

OpenAI · September 23, 2026

In remarks to the Security Council, Altman named two failure modes to avoid: losing control of the future to AI, especially as systems approach recursive self-improvement, and excessive concentration of power. He said OpenAI has unilaterally slowed down before and will again, and called for complementary national and international frontier AI standards covering capability measurement, risk assessment, safeguard sufficiency, incident reporting, and secure channels for sharing emerging threats, without locking in incumbents.

“Beating companies in a competitive pace is not a reason to make rash decisions.” — Sam Altman, OpenAI

Source: Sam Altman’s remarks at the United Nations Security Council

OpenAI extends its Daybreak cyber-defense program to the Government of Ukraine

OpenAI · September 23, 2026

Announced on the sidelines of the UN General Assembly, the arrangement with Ukraine’s Ministry of Digital Transformation gives Ukrainian teams access to OpenAI’s Daybreak tools for finding software vulnerabilities and developing and testing fixes to protect civilian infrastructure. OpenAI noted its cyber models are already used by defenders in France, Germany, and Poland, and by the EU cyber agency ENISA, and that CERT Polska used them to help find six vulnerabilities in third-party router software.

“Ukraine is already on the front line, and its defenders need support now.” — Sasha Baker, Head of National Security Policy, OpenAI

Source: OpenAI extends cyber access to Ukraine for civilian defense

Google adds Live Avatar to Gemini 3.8 Live for real-time, lip-synced video agents

Google · September 24, 2026

Google paired its Gemini 3.8 Live native dialogue model with low-latency streaming video, producing enterprise agents that listen, see, and speak through an animated persona with lip-sync and facial expressions. The feature supports asynchronous tool calling so the avatar keeps talking while fetching data in the background, switches among 97 languages mid-conversation, and can generate custom avatars from a reference image for allowlisted enterprises. All audio and video output is watermarked with SynthID, and the feature is available now in Gemini Enterprise.

“Starting today, Gemini 3.8 Live with Live Avatar is available in Gemini Enterprise.” — Google

Source: Introducing Gemini 3.8 Live with Live Avatar

This brief covers the trailing ~72 hours (September 22–25, 2026).

Primary sources:

Leave a Reply