Claude, the essentials — edition of September 25, 2026
Claude Opus 5.5 Arrives as Anthropic's New Biology Lab Reports Its First Discovery
Anthropic's flagship model gets a major refresh with a 1-million-token context window, while a newly disclosed life sciences lab says Claude helped find a CRISPR-like enzyme system whose function is still unknown.
- Claude Opus 5.5 is now Anthropic's default Opus model, with a 1M-token context window and per-token API pricing of $4/$20 (input/output) plus $0.20 for cached reads.
- Anthropic also introduced "Opus 5.5 for Work," a business-oriented framing of the same release.
- A life sciences research group formed by Anthropic in spring 2026 says Claude autonomously identified a novel, CRISPR-like enzyme system from DNA repeat patterns — but Anthropic states plainly that the system's actual function is not yet known.
- Anthropic published research on measuring the pace of AI development inside frontier labs and on the implications of agents trading on people's behalf.
- Claude Code's latest releases add enterprise controls for AWS Bedrock deployments and change how certain refusals are billed on the API.
A New Flagship: Claude Opus 5.5
Anthropic released Claude Opus 5.5 and made it the default Opus model across its platform. According to Claude Code's release notes, the model ships with a 1-million-token context window and API pricing of $4 per million input tokens and $20 per million output tokens, with cached reads discounted to $0.20 per million tokens. As with prior flagship releases, Anthropic paired the launch with a system card documenting the model's evaluated capabilities and safety testing.
The company also announced "Opus 5.5 for Work," a separate, enterprise-oriented framing of the same model. Publishing a dedicated business-facing announcement alongside the general release suggests Anthropic continues to tailor how it presents a single new model to different audiences rather than issuing one undifferentiated launch.
Sources: Introducing Claude Opus 5.5 — Anthropic · Claude Opus 5.5 System Card — Anthropic · Claude Code 2.1.280 — Added Claude Opus 5.5 (claude-opus-5-5), now the default Opus model — Claude Code · Opus 5.5 for Work — Anthropic
Inside Anthropic's New Biology Lab
Anthropic disclosed the existence of a life sciences research group and laboratory, formed in spring 2026 to test whether general-purpose AI models can systematize and accelerate biological discovery. The company frames this against history: major breakthroughs like restriction enzymes, Taq polymerase, and CRISPR itself all began with scientists noticing something anomalous in nature's molecular machinery, and the new team's working model is one in which AI agents collaborate with human scientists at every step, from mining DNA datasets to running physical lab experiments.
The group's first disclosed result is that Claude autonomously identified a novel enzyme system associated with an array of DNA repeats structurally reminiscent of CRISPR, working from only high-level direction by human scientists. Anthropic is explicit that this is an early and incomplete finding: it states that the discovered system's actual function is not yet known, even though the specific combination of characteristics it displays has previously been observed only in a handful of other systems, all of them programmable tools capable of operations such as cutting DNA.
Sources: Claude discovers a novel enzyme system with CRISPR-like repeats — Anthropic · Autonomous AI agents discover reverse transcriptases with tandem repeat arrays — Anthropic
Measuring Progress and Agentic Risk
In a separate research post, Anthropic addressed how to measure the pace of AI development inside frontier labs, an effort aimed at giving outside observers better signal into how quickly internal capabilities are advancing rather than relying solely on public releases as a proxy.
The company also published "Project Swap," a piece examining what happens when AI agents are delegated the ability to trade on a person's behalf. Taken together with the measurement piece, the two publications reflect a recurring thread in Anthropic's public research this week: scrutinizing both the speed of capability growth and the practical consequences of granting agents real-world economic autonomy.
Sources: Measurements for understanding the pace of AI development inside frontier labs — Anthropic · Project Swap: What happens when agents trade for us? — Anthropic
Developer Platform: Enterprise Gateway Controls and Billing Changes
Claude Code's 2.1.281 and 2.1.282 releases focused on enterprise deployment plumbing. The Claude apps gateway gained assume_role support for Bedrock upstreams, letting the gateway call Bedrock through an IAM role obtained via AWS STS — including in a separate AWS account, optionally with one session per developer — as well as the ability to attach an Amazon Bedrock guardrail (by id and version) to every request on an upstream. The gateway also now recognizes newer Claude Desktop policy keys, including blockReadsOutsideWorkingDirectories and disableBypassPermissionsMode.
The same releases brought smaller usability changes: a maxProseWidth setting that caps prose width in wide terminals while leaving tables and code blocks at full width, and startup, /status, and claude doctor notices that surface when a project's settings are silently ignoring or disabling telemetry variables. On the API side, Anthropic's release notes recorded two further changes: cache diagnostics graduated out of beta and no longer require a beta header, and the company began billing certain refusals that arrive before any output — specifically those flagged under the "bio," "frontier_llm," or "reasoning_extraction" categories, where measured false-positive rates are low — bringing them in line with mid-stream refusals, which were already billed.
Sources: Claude Code 2.1.282 release notes — Claude Code · Claude Code 2.1.281 release notes — Claude Code · API release notes — September 24, 2026 — Anthropic (release notes) · API release notes — September 23, 2026 — Anthropic (release notes)
This edition is an original synthesis written by Claude from aggregated news — Anthropic's own sources first (release notes, status, newsroom, research, engineering), then the press, Hacker News, Reddit and GitHub, under the editorial supervision of Héra SASU. Every fact links to its article, publisher named. See the live feed →
Claude News is published by Héra SASU. Independent media, not affiliated with Anthropic.