Neon backend goes GA and starts billing, three tools drop legacy APIs, Pinecone auto-routes around Gemini/o4-mini deprecations
The radar watched 63 sources across 32 tools this window and published 28 changes; these are the twelve worth your attention, ranked. That's two more than the typical ten: three separate tools, Langfuse, the Vercel AI SDK, and Weaviate, removed legacy APIs in the same week, and Neon's backend GA landed alongside the pricing change that ends its free beta, so both halves of that story stayed in rather than cutting one. Every item carries the before/after evidence the pipeline captured; expand it before you take our word.
Neon's backend is now generally available, expanding from serverless Postgres into a full backend platform where every branch is a complete copy of the stack, not just the database. Object Storage, Functions, and AI Gateway move out of free beta at the same time (see the pricing change below), so budget for them before scaling up.
featurehigh
Neon backend reaches general availability (GA)
Neon users can now use the full backend service in production with all branching capabilities, expanding from Postgres-only offering to complete backend infrastructure.
Evidence
+The Neon backend is generally available release - Sep 18, 2026
The Neon backend is generally available. Neon started as serverless Postgres. Now you get the whole backend, and all of it branches. Every branch is a full copy...
Following the GA above, Neon starts charging for Object Storage ($0.023 per GB-month), Functions (compute and invocations), and AI Gateway (prepaid credits at provider list price, no markup). Anyone who built on these during the free beta needs to re-check their bill.
pricinghigh
Object Storage, Functions, and AI Gateway billing activated; previously free during beta
Users will now be charged for these services. Object Storage costs $0.023 per GB-month; Functions are billed for compute and invocations; AI Gateway draws down prepaid credits.
Evidence
−No charges during beta
ChargesNo charges during beta
They aren't billed yet. Usage is free of charge for all three during beta
+$0.023 per GB-month
List prices herePrepaid credits
Object Storage is $0.023 per GB-month of stored volume. Functions are billed for active compute, waiting compute, and invocations at the rates in the plan table. AI Gateway is available on Launch and Scale. Pricing matches the model provider's list prices, with no markup, and draws down prepaid credits.
Google deprecated Gemini 2.5 Pro and OpenAI deprecated o4-mini; Pinecone Assistant now silently routes any request naming those models to Gemini 3.5 Flash or GPT-5 instead, at the same price. Convenient, but it also means your Assistant calls may already be running on a different model than the one in your code.
deprecationhigh
Gemini 2.5 Pro and o4-mini models deprecated; automatic routing to newer models implemented
Existing code using gemini-2.5-pro will automatically route to Gemini 3.5 Flash, and o4-mini requests will route to GPT-5 at the same price. Users are encouraged to update code explicitly but no changes are required.
Evidence
+Gemini 2.5 Pro and o4-mini deprecations for Assistant
Google has deprecated the Gemini 2.5 Pro model, and OpenAI has deprecated o4-mini. Pinecone Assistant automatically routes all chat requests that specify gemini-2.5-pro to Gemini 3.5 Flash, and all chat requests that specify o4-mini to GPT-5, at the same price.
@ai-sdk/xai 5.0.0 drops the Chat Completions API entirely, keeping only the newer Responses-style interface. If you're on an older xai integration, this upgrade breaks it rather than just deprecating it.
Langfuse now rejects legacy GET-based reads for any newly created organization; existing orgs keep the old behavior for now. New projects should build against the current read API from day one rather than copying patterns from an older integration.
apihigh
API now rejects legacy GET reads for new organizations
New organizations will no longer support legacy GET read operations; existing integrations using legacy reads will fail and need migration.
Evidence
+feat(api)!: reject legacy reads for new organizations
Weaviate removed the /v1/modules REST endpoint along with the contextionary extension and concept handlers. If any tooling still queries /v1/modules or relies on contextionary concept expansion, it stops working on upgrade, not just prints a deprecation warning.
deprecation
Removal of /v1/modules endpoint and contextionary extension/concept handlers
Users relying on the /v1/modules endpoint, contextionary extensions, or concept handlers will need to update their integration code. This API endpoint is no longer available.
Evidence
+modules(rest): remove /v1/modules endpoint and contextionary extension/concept handlers
Z.ai introduced GLM-5.3-FlashX at $0.37 and $0.075 per million tokens, with a limited-time free tier and $1.25 batch pricing. Check the pricing page directly for which rate applies to input versus output before committing a workload to it.
pricinghigh
Introduction of GLM-5.3-FlashX model with tiered pricing structure
Developers can now use the new GLM-5.3-FlashX model with input pricing at $0.37, output pricing at $0.075, limited-time free tier option, and batch pricing at $1.25
DeepSeek-Flash now bills separately for off-peak and peak hours: cache-hit input runs $0.003 off-peak versus $0.006 peak, cache-miss input $0.15 versus $0.3, output $0.6 versus $1.2 per million tokens. Workloads that can shift to off-peak hours get roughly half price.
pricinghigh
DeepSeek-Flash pricing structure introduced with off-peak and peak rates ($0.003-$1.2 per 1M tokens)
Users can now see transparent pricing for deepseek-flash with time-based pricing variations, cache hit discounts, and off-peak/peak rate differentiation.
Kimi's new K3 series models introduce two cache TTL tiers, 5 minutes and 1 hour (5 minutes is the default when unspecified). Cache writes are billed separately per tier, but a cache hit is billed only at the cached-input rate with no extra write charge, and each hit refreshes the entry's lifetime.
featurehigh
Introduced K3 Series Models with context cache pricing structure
Users can now use K3 Series models with automatic request prefix caching (5min and 1h TTL options), enabling cost savings through cached input billing at lower rates than regular input.
Evidence
+K3 Series Models
Note: Cache Write refers to the cost of writing request prefixes into the context cache. The Kimi API automatically caches repeated request prefixes, and cache entries come in two time-to-live (TTL) tiers — 5min and 1h. If no TTL is specified, the 5min tier applies by default. Input that hits the cache within its lifetime is billed only at the Cached Input Price: each hit automatically refreshes the cache entry's lifetime, with no additional cache write charge.
OpenAI added API key creation governance at the organization and project level: admins can restrict new keys to service accounts only, to user-owned project keys only, or block new key creation entirely. Existing keys keep working, and org-level restrictions override project settings.
featurehigh
Added API key governance controls at organization and project levels
Administrators can now enforce stricter API key management policies, restricting key creation to service accounts or project keys, and existing keys remain unaffected. This improves security and compliance for organizations.
Evidence
+Added API key creation governance controls at the organization and project levels. Administrators can allow only service-account keys, allow only user-owned project keys, or disable all new API key creation. Organization restrictions take precedence over project settings, and existing API keys are unaffected.
Claude Code now reads AGENTS.md instead of CLAUDE.md when a project has no CLAUDE.md, toggled under Project instructions in /config. Useful if you already maintain an AGENTS.md for other agent tools and don't want to duplicate it, though it's not yet available on Bedrock, Vertex, or Foundry.
feature
Added AGENTS.md support as alternative to CLAUDE.md for project instructions
Users can now use AGENTS.md instead of CLAUDE.md for project instructions in Claude Code, configurable via /config panel (not yet available on Bedrock, Vertex, or Foundry)
Evidence
+Added AGENTS.md support: in a project with no CLAUDE.md, Claude Code reads AGENTS.md instead; change it under "Project instructions" in /config (not yet on Bedrock, Vertex or Foundry)
GitHub Copilot's budget increase request flow is now generally available: when a member hits their AI credit limit, they can request more instead of just getting blocked, and org or enterprise owners approve, adjust, or deny the request from settings. Worth turning on before your team's heaviest Copilot users start hitting the wall silently.
featurehigh
Copilot budget increase requests workflow now generally available
Organization and enterprise owners can now approve, adjust, or deny member budget requests directly in settings. Members can request more AI credits when they hit their limit instead of being blocked from Copilot features.
Evidence
+Previously, when a member used all the Copilot AI credits available to them, they were blocked from Copilot features that consume credits. This release adds a flow for them to request more budget the moment they hit the limit. You can approve, adjust, or deny that request without leaving your settings.