Claude Haiku 5.5: Pricing, Token Thresholds and Migration Changes
Decide whether to migrate from Haiku 4.5: compare two-tier prices, recount prompts and check thinking, tool use, refusals and stored-session behavior.
Blog
Notes on building reliable AI products.
GPT-6 Brings Intelligent UI to ChatGPT: Rollout and API Guidance
GPT-6 adds interactive answers to ChatGPT. See the rollout dates, Pro reasoning exception and why API model selection remains a separate decision.
Codex CLI 0.161.0: Bedrock, MCP Login and Cyber Upgrade Guide
When to upgrade Codex CLI 0.161.0: check Bedrock routing and GovCloud limits, in-session MCP login, and opt-in Cyber program controls.
SynthID Detector Goes Public: What Media Checks Can and Cannot Prove
Google’s public SynthID Detector checks images, video and audio. Learn how to interpret results, pair them with C2PA and assess upload and automation limits.
Nemotron IOI and IMO: Models, Datasets and How to Start
Choose NVIDIA’s Nemotron IOI and IMO artifacts, check compute and context requirements, and plan a bounded proof-workflow run without overstating the scores.
Google Playground: Creation Access, Source Export and Sharing Limits
Google Playground offers prompt-based browser games for U.S. adults. Compare creation access, source export, sharing limits and the forthcoming Unity Spark integration.
Anthropic Project Swap: Test User Preferences Before Agents Negotiate
What Anthropic’s book-trading experiment reveals about preference capture, negotiation metrics and the checks to run before delegating exchanges to AI agents.
OpenAI API Moves to Build, Launch and Grow Usage Tiers
OpenAI replaces five paid API tiers with Build, Launch and Grow. Compare credit thresholds, monthly ceilings and model rate limits before a launch.
OpenAI Decisions API Beta: When to Use It Instead of Responses
Choose between OpenAI Decisions and Responses for support routing and classification, with request constraints, fallback costs and evaluation guidance.
EmbeddingGemma 2: A Builder’s Guide to Local Multimodal Search
Choose EmbeddingGemma 2 encoders and vector dimensions, connect local media search, and account for sampling limits before replacing a retrieval pipeline.
Anthropic’s Cyber Verification Program: Choose a Tier and Check Retention
Compare Anthropic’s expanded CVP tiers, security requirements, cloud access and retention rules before applying for Claude cyber capabilities.
OpenAI–Ironclad Contracting Evals: What a Workflow Pilot Must Prove
OpenAI’s Ironclad research informs a contracting-agent pilot: test approval routing, reapproval and configuration behavior before deployment.