Daily Pulse

At least 5 verified AI/ML items every day, each with a "so what". Every card opens into a full release page, so you choose how deep to go.

Provider

Tier

Since

My Stack

ModelsAnthropic23 Sept 2026

ConfirmedFrontier model release · Tier 1

Claude Opus 5.5 ships: Opus-tier quality at a 20% lower price

So what? Teams on Claude Opus 5 get a cheaper successor with the same 1M context and 128K output. Four breaking API changes (thinking can't be disabled, lower default effort, no forced tool_choice, preserved thinking) mean this is a code change, not a model-string swap.

ImpactWhereDirect API global · AWS Bedrock us-east-1, us-west-2, ap-south-1 (preview) · Google Vertex AI global · Azure AI Foundry eastus2 (preview)SourcesAnthropic announcement, Models overview (docs), Pricing

Full release page →4 proofs 2 POCs

Frameworks & ToolsColibrì23 Sept 2026

ConfirmedFramework or tooling · Tier 2

Colibrì runs the 744B GLM-5.2 on 16–25 GB of RAM with no GPU, slowly

So what? A tiny open-source C engine streams a frontier-size MoE model's experts from disk, so it runs offline with no GPU and no per-token bill. Needs ~372 GB of NVMe and runs at about 1 token every 10–20 s on a 25 GB machine: great for private batch jobs, not for chat.

ImpactSourcesColibrì on GitHub, GLM-5.2 int4 weights (Hugging Face), Viral Short (lead) (lead)

Read the claim check →

Cloud PlatformsAWS Bedrock & SageMaker23 Sept 2026

ConfirmedCloud AI platform change · Tier 1

Claude Opus 5.5 reaches AWS Bedrock in Mumbai (ap-south-1), preview

So what? Indian teams with data-residency requirements can trial Opus 5.5 in-region the day after launch. Preview means no SLA and no batch inference yet, so keep production on Opus 5 in ap-south-1 for now.

ImpactWhereDirect API global · AWS Bedrock us-east-1, us-west-2, ap-south-1 (preview) · Google Vertex AI global · Azure AI Foundry eastus2 (preview)SourcesAWS What's New, Bedrock model support by region

Full release page →4 proofs 2 POCs

Frameworks & ToolsLangGraph23 Sept 2026

ConfirmedFramework or tooling · Tier 2

LangGraph (demo release) renames checkpointer interfaces — breaking for custom savers

So what? If you wrote a custom checkpointer (for example on pgvector/Postgres), it will not load after upgrading. Pin the previous minor version until you port the interface.

ImpactSourcesGitHub release notes

ModelsAnthropic23 Sept 2026

ConfirmedAPI and feature change · Tier 2

Claude API: per-message effort control (beta) without a prompt-cache reset

So what? Agent loops can drop to low effort for routine turns and raise it for hard ones mid-conversation while keeping the cached prefix. Useful cost lever for long-running agents.

ImpactSourcesClaude API release notes

ResearcharXiv (research to practice)23 Sept 2026

ConfirmedResearch to practice · Tier 2

Paper with code (demo): late-interaction reranking cuts RAG hallucinations on long PDFs

So what? A runnable reranker that drops into existing retrieval pipelines. Worth a 30-minute test if your RAG answers cite the wrong page; our RAG suite run is scheduled.

ImpactSourcesarXiv listing (demo)

ModelsMeta23 Sept 2026

ConfirmedBonus · Tier 2

Llama open-weight 2 70B (demo) lands on Bedrock in preview

So what? A cheap open-weight option for high-volume classification on Bedrock. No structured-output support yet, so extraction pipelines need a validator.

ImpactWhereAWS Bedrock us-east-1 (preview)SourcesVendor blog (demo)

Model page

ModelsOpenAI23 Sept 2026

RumorBonus · Tier 2

Rumor: OpenAI Frontier 3 (demo) in limited testing

So what? Nothing to do yet. We track it so you are not surprised; no facts will be added until a primary source confirms.

ImpactSourcesHacker News thread (early signal only) (lead)

ModelsGoogle DeepMind22 Sept 2026

Announced (not yet available)Frontier model release · Tier 1

Gemini Pro 4 (demo) announced for 1 October with 2M context

So what? Long-context workloads may get cheaper per document. Not available yet; plan an eval, don't plan a migration.

ImpactWhereNot yet availableSourcesVendor blog (demo)

Model page

Cloud PlatformsAzure AI Foundry / Azure OpenAI22 Sept 2026

ConfirmedCloud AI platform change · Tier 1

OpenAI Frontier 2 (demo) in preview on Azure Central India

So what? In-region option for Azure-first Indian teams. Preview quotas are small; request an increase before load tests.

ImpactWhereDirect API global · Azure AI Foundry eastus2, centralindia (preview)SourcesAzure AI Foundry what's new

Full release page →1 POC

Frameworks & ToolsvLLM22 Sept 2026

ConfirmedFramework or tooling · Tier 2

vLLM (demo release) adds speculative decoding for open-weight 70B models

So what? Self-hosters can get higher tokens/sec on the same GPUs. Benchmark on your own prompts; gains depend on output length.

ImpactSourcesGitHub release notes

Frameworks & ToolsModel Context Protocol22 Sept 2026

ConfirmedAPI and feature change · Tier 2

MCP spec revision (demo) adds resumable streams

So what? Long-running tool calls survive dropped connections. MCP server authors should plan an upgrade; clients are backward compatible.

ImpactSourcesMCP changelog

Frameworks & ToolsOllama19 Sept 2026

ConfirmedFramework or tooling · Tier 3

Minor releases: Ollama, CrewAI and pgvector patch versions (demo)

So what? Bug fixes only. Upgrade on your normal schedule.

ImpactSourcesGitHub releases

Cloud PlatformsGoogle Vertex AI15 Sept 2026

ConfirmedCloud AI platform change · Tier 1

Gemini Pro 3 (demo) price cut: input -20%, output -20%

So what? Now cheaper than several peers on our RAG suite at similar accuracy. Re-run the cost calculator if you chose a model on price in the last quarter.

ImpactWhereDirect API global · Google Vertex AI asia-south1SourcesVertex AI pricing

Model page 1 proof 1 POC

ModelsOpenAI28 Aug 2026

ConfirmedFrontier model release · Tier 1

OpenAI Frontier 2 (demo) released; Frontier 1 deprecated for 1 December

So what? Frontier 1 users have ~3 months to migrate. The new version adds audio input and a 1M context window.

ImpactWhereDirect API global · Azure AI Foundry eastus2, centralindia (preview)SourcesVendor deprecations page (demo)

Full release page →1 POC