SShortSingh.
TopTechnologyAI & MLStartupsProgrammingScienceBusinessFinanceCrypto & Web3World
Back to feed
Programming

Claude Opus 5.5 cuts costs 40%, delivers faster output with no API changes

0
·1 views

Anthropic released Claude Opus 5.5 this week with a 40% price reduction and 30% faster output generation compared to its predecessor, while matching Fable 5.1 benchmark performance. The most significant saving is in cache reads, which drop from $0.50 to $0.20 per million tokens — a 60% reduction that benefits agentic and retrieval-heavy workflows. No API or prompt changes are required; developers can migrate simply by updating the model string or staying on the latest alias. Separately, Vercel's AI Gateway expanded support to include four new models — GLM-5.3 Flash, DeepSeek V4.1 Flash, Qwen 3.8 Flash, and Grok 4.7 — consolidating provider access under a single integration layer. Google's Gemini 3.5 Transcribe also joined the Gateway, offering WebSocket-based live transcription across 85+ languages without requiring a separate Google Speech-to-Text credential setup.

Read the full story at DEV Community

This is an AI-generated summary. ShortSingh links to the original source for the complete article.

Discussion (0)

Log in to join the discussion and vote.

Log in

Related stories

0
ProgrammingDEV Community ·

AWS S3 vs EBS vs EFS: Key Differences and When to Use Each

Amazon Web Services offers three distinct storage solutions — S3, EBS, and EFS — each designed for fundamentally different use cases. EBS functions as a virtual hard drive attached to a single EC2 instance, making it suitable for operating systems and databases requiring low-latency block access. EFS is a managed network file system that multiple instances can mount simultaneously, ideal for shared workloads across servers, though it costs more per GB than EBS. S3 is object storage accessed via HTTP API rather than a file system, best suited for backups, static assets, logs, and data lakes at a cheaper per-GB cost. Choosing the wrong service is a common beginner mistake that can result in inefficient architectures and higher-than-expected cloud bills.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

In-Process PII Scrubbing Tool 'The Cleaner' Aims to Stop LLM Data Leaks

Developers building AI agents risk exposing sensitive user data — such as credit card numbers, IBANs, and personal emails — by passing raw database records directly into large language model contexts. Once that data leaves the server, it can end up in third-party provider logs, inference caches, or even be echoed back in public-facing chat responses. A proposed solution called 'The Cleaner', part of the avantGate security framework, runs in-process to mask PII before any payload crosses the network boundary to an external LLM. It operates on a zero-trust principle, intercepting inputs at ingress and enforcing a dual-channel tool boundary that sends restricted data to the model while routing full payloads directly to the client. The approach is designed to help teams meet compliance requirements under GDPR, HIPAA, and PCI-DSS without adding significant network latency.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

OpenAI Rates GPT-6 Astra Critical for Cybersecurity Risk Over Zero-Day Exploit Ability

OpenAI has assigned its first-ever Critical cybersecurity rating to GPT-6 Astra, a model capable of autonomously discovering zero-day vulnerabilities and building working exploits. Although the capability is currently gated, security experts warn that similar tools could reach malicious actors sooner than anticipated. In response, security practitioners are advising organisations to audit their cloud environments by mapping all publicly exposed assets — including IPs, S3 buckets, load balancers, and admin panels — and eliminating unnecessary exposure. Teams are also urged to measure and reduce patch latency specifically for internet-facing systems, as faster exploit development shrinks the window between vulnerability disclosure and active exploitation. Additional recommended steps include enforcing least-privilege IAM permissions on exposed workloads and deploying independent anomaly detection to catch signs of compromise early.

0 comments Read more at DEV Community
0
ProgrammingDEV Community ·

Claude Code Subagents Ate 48% of One Developer's Bill While Producing 0.9% of Output

A developer analyzing a month of Claude Code usage discovered that AI subagents accounted for 48% of total costs despite generating only 0.9% of output tokens. The disparity stems from how agent loops work: every tool call resends the full context window, including system prompts, tool schemas, memory files, and prior conversation history. In the developer's setup, each subagent began with roughly 51,000 tokens of fixed overhead, which was re-sent on every single model request within the loop. Long main sessions compounded the problem, with 45 sessions exceeding $100 each and accounting for 79% of total spend. The developer now recommends capping agents per run, batching smaller tasks into a single agent, trimming starting context, and handing off to a fresh session before context grows too large.

0 comments Read more at DEV Community