{"generatedFrom":"aimluptodate.com","count":15,"items":[{"id":"2026-09-23-claude-opus-5-5","date":"2026-09-23","headline":"Claude Opus 5.5 ships: Opus-tier quality at a 20% lower price","status":"confirmed","soWhat":"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.","impact":5,"impactReason":"Frontier model, direct price cut for every Opus workload, and breaking API changes.","slot":"frontier-release","tier":1,"providerId":"anthropic","category":"models","tags":["claude","anthropic","bedrock","vertex","azure","pricing","breaking-change"],"versionId":"claude-opus-5-5","sources":[{"label":"Anthropic announcement","url":"https://www.anthropic.com/news","kind":"primary"},{"label":"Models overview (docs)","url":"https://docs.claude.com/en/docs/about-claude/models/overview","kind":"primary"},{"label":"Pricing","url":"https://www.anthropic.com/pricing","kind":"primary"}],"verdict":{"summary":"Opus 5.5 replaces Opus 5 at $4 / $20 per 1M tokens (vs $5 / $25), same context and output limits.","whoFor":"Anyone running Opus 5 in production; teams choosing a frontier model for coding and agents.","shouldYouCare":"Yes if you run Opus 5: ~20% cheaper. Budget a day for the four breaking changes before switching."},"videos":[{"title":"Opus 5.5 migration walkthrough (demo link)","url":"https://www.youtube.com/","contributorId":"c-ravi"}],"provider":"Anthropic","permalink":"https://aimluptodate.com/pulse/2026-09-23-claude-opus-5-5/","releasePage":"https://aimluptodate.com/releases/claude-opus-5-5/"},{"id":"2026-09-23-colibri-744b-no-gpu","date":"2026-09-23","headline":"Colibrì runs the 744B GLM-5.2 on 16–25 GB of RAM with no GPU, slowly","status":"confirmed","soWhat":"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.","impact":4,"impactReason":"Changes what hardware can run frontier-size open models; too slow for interactive use.","slot":"framework-tooling","tier":2,"providerId":"colibri","category":"frameworks","tags":["open-weight","local-ai","offline","glm","moe","no-gpu"],"sources":[{"label":"Colibrì on GitHub","url":"https://github.com/JustVugg/colibri","kind":"primary"},{"label":"GLM-5.2 int4 weights (Hugging Face)","url":"https://huggingface.co/mastouri/GLM-5.2-colibri-int4-g64-with-int8-mtp","kind":"primary"},{"label":"Viral Short (lead)","url":"https://www.youtube.com/shorts/WYJxGSfXLZQ","kind":"lead"}],"provider":"Colibrì","permalink":"https://aimluptodate.com/pulse/2026-09-23-colibri-744b-no-gpu/"},{"id":"2026-09-23-bedrock-opus-5-5-mumbai","date":"2026-09-23","headline":"Claude Opus 5.5 reaches AWS Bedrock in Mumbai (ap-south-1), preview","status":"confirmed","soWhat":"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.","impact":4,"impactReason":"In-region availability unblocks regulated workloads; preview limits production use.","slot":"cloud-platform","tier":1,"providerId":"aws","category":"cloud","tags":["bedrock","ap-south-1","claude","region"],"versionId":"claude-opus-5-5","sources":[{"label":"AWS What's New","url":"https://aws.amazon.com/about-aws/whats-new/","kind":"primary"},{"label":"Bedrock model support by region","url":"https://docs.aws.amazon.com/bedrock/latest/userguide/models-regions.html","kind":"primary"}],"provider":"AWS Bedrock & SageMaker","permalink":"https://aimluptodate.com/pulse/2026-09-23-bedrock-opus-5-5-mumbai/"},{"id":"2026-09-23-langgraph-demo-breaking","date":"2026-09-23","headline":"LangGraph (demo release) renames checkpointer interfaces — breaking for custom savers","status":"confirmed","soWhat":"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.","impact":3,"impactReason":"Breaking change for a subset of production agent deployments.","slot":"framework-tooling","tier":2,"providerId":"langgraph","category":"frameworks","tags":["langgraph","agents","breaking-change","pgvector"],"sources":[{"label":"GitHub release notes","url":"https://github.com/langchain-ai/langgraph/releases","kind":"primary"}],"provider":"LangGraph","permalink":"https://aimluptodate.com/pulse/2026-09-23-langgraph-demo-breaking/"},{"id":"2026-09-23-claude-api-per-message-effort","date":"2026-09-23","headline":"Claude API: per-message effort control (beta) without a prompt-cache reset","status":"confirmed","soWhat":"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.","impact":3,"impactReason":"Direct cost lever for agent workloads; beta only.","slot":"api-feature","tier":2,"providerId":"anthropic","category":"models","tags":["claude","anthropic","api","caching","cost"],"sources":[{"label":"Claude API release notes","url":"https://docs.claude.com/en/release-notes/api","kind":"primary"}],"provider":"Anthropic","permalink":"https://aimluptodate.com/pulse/2026-09-23-claude-api-per-message-effort/"},{"id":"2026-09-23-demo-rag-paper","date":"2026-09-23","headline":"Paper with code (demo): late-interaction reranking cuts RAG hallucinations on long PDFs","status":"confirmed","soWhat":"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.","impact":2,"impactReason":"Practical technique with code; not yet measured by us.","slot":"research-practice","tier":2,"providerId":"arxiv","category":"research","tags":["rag","reranker","research"],"sources":[{"label":"arXiv listing (demo)","url":"https://arxiv.org/list/cs.CL/recent","kind":"primary"}],"provider":"arXiv (research to practice)","permalink":"https://aimluptodate.com/pulse/2026-09-23-demo-rag-paper/"},{"id":"2026-09-23-demo-llama-open-2","date":"2026-09-23","headline":"Llama open-weight 2 70B (demo) lands on Bedrock in preview","status":"confirmed","soWhat":"A cheap open-weight option for high-volume classification on Bedrock. No structured-output support yet, so extraction pipelines need a validator.","impact":2,"impactReason":"Useful cost option; limited features.","slot":"bonus","tier":2,"providerId":"meta","category":"models","tags":["open-weight","bedrock","llama"],"versionId":"demo-llama-open-2","sources":[{"label":"Vendor blog (demo)","url":"https://ai.meta.com/blog/","kind":"primary"}],"provider":"Meta","permalink":"https://aimluptodate.com/pulse/2026-09-23-demo-llama-open-2/"},{"id":"2026-09-23-demo-openai-rumor","date":"2026-09-23","headline":"Rumor: OpenAI Frontier 3 (demo) in limited testing","status":"rumor","soWhat":"Nothing to do yet. We track it so you are not surprised; no facts will be added until a primary source confirms.","impact":1,"impactReason":"Unconfirmed; no action possible.","slot":"bonus","tier":2,"providerId":"openai","category":"models","tags":["openai","rumor"],"sources":[{"label":"Hacker News thread (early signal only)","url":"https://news.ycombinator.com/","kind":"lead"}],"provider":"OpenAI","permalink":"https://aimluptodate.com/pulse/2026-09-23-demo-openai-rumor/"},{"id":"2026-09-22-demo-gemini-pro-4-announced","date":"2026-09-22","headline":"Gemini Pro 4 (demo) announced for 1 October with 2M context","status":"announced","soWhat":"Long-context workloads may get cheaper per document. Not available yet; plan an eval, don't plan a migration.","impact":3,"impactReason":"Large context jump, but not yet available.","slot":"frontier-release","tier":1,"providerId":"google-deepmind","category":"models","tags":["gemini","vertex","long-context"],"versionId":"demo-gemini-pro-4","sources":[{"label":"Vendor blog (demo)","url":"https://deepmind.google/discover/blog/","kind":"primary"}],"provider":"Google DeepMind","permalink":"https://aimluptodate.com/pulse/2026-09-22-demo-gemini-pro-4-announced/"},{"id":"2026-09-22-azure-demo-region","date":"2026-09-22","headline":"OpenAI Frontier 2 (demo) in preview on Azure Central India","status":"confirmed","soWhat":"In-region option for Azure-first Indian teams. Preview quotas are small; request an increase before load tests.","impact":3,"impactReason":"New region for a frontier model.","slot":"cloud-platform","tier":1,"providerId":"azure","category":"cloud","tags":["azure","centralindia","region","openai"],"versionId":"demo-openai-frontier-2","sources":[{"label":"Azure AI Foundry what's new","url":"https://learn.microsoft.com/azure/ai-foundry/whats-new","kind":"primary"}],"provider":"Azure AI Foundry / Azure OpenAI","permalink":"https://aimluptodate.com/pulse/2026-09-22-azure-demo-region/"},{"id":"2026-09-22-vllm-demo","date":"2026-09-22","headline":"vLLM (demo release) adds speculative decoding for open-weight 70B models","status":"confirmed","soWhat":"Self-hosters can get higher tokens/sec on the same GPUs. Benchmark on your own prompts; gains depend on output length.","impact":2,"impactReason":"Throughput gain for self-hosted inference.","slot":"framework-tooling","tier":2,"providerId":"vllm","category":"frameworks","tags":["vllm","inference","open-weight"],"sources":[{"label":"GitHub release notes","url":"https://github.com/vllm-project/vllm/releases","kind":"primary"}],"provider":"vLLM","permalink":"https://aimluptodate.com/pulse/2026-09-22-vllm-demo/"},{"id":"2026-09-22-mcp-demo-spec","date":"2026-09-22","headline":"MCP spec revision (demo) adds resumable streams","status":"confirmed","soWhat":"Long-running tool calls survive dropped connections. MCP server authors should plan an upgrade; clients are backward compatible.","impact":2,"impactReason":"Reliability improvement for agent tooling.","slot":"api-feature","tier":2,"providerId":"mcp","category":"frameworks","tags":["mcp","agents","protocol"],"sources":[{"label":"MCP changelog","url":"https://modelcontextprotocol.io","kind":"primary"}],"provider":"Model Context Protocol","permalink":"https://aimluptodate.com/pulse/2026-09-22-mcp-demo-spec/"},{"id":"2026-09-01-weekly-tools-roundup","date":"2026-09-19","headline":"Minor releases: Ollama, CrewAI and pgvector patch versions (demo)","status":"confirmed","soWhat":"Bug fixes only. Upgrade on your normal schedule.","impact":1,"impactReason":"Patch releases.","slot":"framework-tooling","tier":3,"providerId":"ollama","category":"frameworks","tags":["ollama","crewai","pgvector"],"sources":[{"label":"GitHub releases","url":"https://github.com/ollama/ollama/releases","kind":"primary"}],"provider":"Ollama","permalink":"https://aimluptodate.com/pulse/2026-09-01-weekly-tools-roundup/"},{"id":"2026-09-15-demo-gemini-price-cut","date":"2026-09-15","headline":"Gemini Pro 3 (demo) price cut: input -20%, output -20%","status":"confirmed","soWhat":"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.","impact":4,"impactReason":"Direct price change on a widely used model.","slot":"cloud-platform","tier":1,"providerId":"gcp","category":"cloud","tags":["gemini","vertex","pricing"],"versionId":"demo-gemini-pro-3","sources":[{"label":"Vertex AI pricing","url":"https://cloud.google.com/vertex-ai/generative-ai/pricing","kind":"primary"}],"provider":"Google Vertex AI","permalink":"https://aimluptodate.com/pulse/2026-09-15-demo-gemini-price-cut/"},{"id":"2026-08-28-demo-openai-frontier-2","date":"2026-08-28","headline":"OpenAI Frontier 2 (demo) released; Frontier 1 deprecated for 1 December","status":"confirmed","soWhat":"Frontier 1 users have ~3 months to migrate. The new version adds audio input and a 1M context window.","impact":5,"impactReason":"Frontier release plus a hard deprecation date.","slot":"frontier-release","tier":1,"providerId":"openai","category":"models","tags":["openai","deprecation","azure"],"versionId":"demo-openai-frontier-2","sources":[{"label":"Vendor deprecations page (demo)","url":"https://platform.openai.com/docs/deprecations","kind":"primary"}],"verdict":{"summary":"Frontier 2 (demo) supersedes Frontier 1 with larger context and audio input; Frontier 1 is deprecated.","whoFor":"Anyone on Frontier 1.","shouldYouCare":"Yes, migration is mandatory by 1 December."},"provider":"OpenAI","permalink":"https://aimluptodate.com/pulse/2026-08-28-demo-openai-frontier-2/","releasePage":"https://aimluptodate.com/releases/demo-openai-frontier-2/"}]}