Sonnet 5.5 task costs, cache prices and API changes
Current Sonnet 5.5 docs list lower cache-read pricing. Review thinking and tool-call changes, then compare the full cost of an accepted task.
Articles
Guides, news and explanations about AI and reliable systems.
Not sure where to begin? Choose a short reading path, or browse at your own pace.
Codex Cloud reuses the setup while each task keeps its work
Publish a prepared toolchain for new coding tasks. A resumed task keeps its own files and tools, so a republished environment can leave old and new tasks different.
Google ends maintenance for legacy Monitoring and Logging agents
Google’s legacy agents have left standard maintenance. During replacement, a small change to a disk’s name can leave a healthy-looking system with an empty alert.
CLM-8B lets agents rank a supplied set of actions
CLM can save the work of comparing a ticket with a stable team directory. The useful test is whether the ticket reaches the right person, including when none of the choices fits.
AlloyDB gives agent queries their own compute in preview
Read-only assistants can query current data on separate compute. Test the wait after idle time alongside the application traffic that still uses shared storage.
Google PageBreak tests AI-found bugs before sending reports
Google’s internal PageBreak project tests suspected flaws before sending them to developers. The useful result is a failure someone can reproduce and a repair they can check.
Storage Intelligence adds usage findings and batch dry runs
Storage Intelligence flags unusual storage usage and previews batch changes. A recurring dataset read shows why the job’s purpose matters when choosing which objects to move.
CloudWatch Omni brings agent evaluations and telemetry into one workspace
A shared screen can connect an agent’s failed call with the service behind it. CloudWatch Omni still needs the right records, or a dependable way to reach them.
A blank error-rate graph needs an explanation
Keep zero errors, no traffic and missing measurements distinct in Prometheus before a dashboard or release gate turns a gap into reassurance.
Claude Opus 5.5 costs less; test whether it keeps the exception
Compare Opus 5.5 on the same operational decision and client behavior, including the one condition that should stop a familiar restart.
NVIDIA’s Cluster Readiness Engine tests GPU workloads before admission
NVIDIA’s Cluster Readiness Engine runs diagnostic, communication and training workloads on selected nodes. Workload-specific thresholds turn those results into an admission decision.
Antigravity brings local models to its agent SDK
Antigravity’s SDK supports Gemma through LiteRT and compatible local model servers. A hybrid cloud planner offers another arrangement, with filenames and task descriptions sent off-device.