GitHub Copilot retires four models, Claude Code defaults to Fable 5.1, Braintrust cuts included credits
The radar watched 63 sources across 32 tools this window and published 29 changes; these are the thirteen worth your attention, ranked. That's three more than the typical ten: five of them (Kimi's Moonshot V1 sunset, its new Anthropic API format support, Mistral's OCR 4.1 GA, and two Pinecone pricing changes) were detected right at last week's window boundary and only got reviewed now, on top of a genuinely dense week for pricing and deprecation news on its own — Claude Code's new default Fable model, a Copilot model cull, and Langfuse cutting session lifetime. Every item carries the before/after evidence the pipeline captured; expand it before you take our word.
GitHub Copilot will deprecate Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, and Claude Opus 4.7 across every surface (Chat, inline edits, agent mode, completions) on October 2. Suggested replacements: Gemini 3.8 Flash for both Flash models, Kimi K3, and Claude Opus 5. If you pin a model in Copilot settings, check now, not on October 2.
deprecationhigh
Multiple GitHub Copilot models deprecated on October 2, 2026
Users relying on Gemini 3.5 Flash, Gemini 3.6 Flash, Kimi K2.7 Code, or Claude Opus 4.7 must migrate to suggested alternatives before October 2, 2026. Administrators need to enable alternative models in Copilot settings.
Evidence
+We will deprecate the following models across all GitHub Copilot experiences (including Copilot Chat, inline edits, ask and agent modes, and code completions) on October 2nd, 2026:
Model | Deprecation date | Suggested alternative
Gemini 3.5 Flash | 2026-10-02 | Gemini 3.8 Flash
Gemini 3.6 Flash | 2026-10-02 | Gemini 3.8 Flash
Kimi K2.7 Code | 2026-10-02 | Kimi K3
Claude Opus 4.7 | 2026-10-02 | Claude Opus 5
Claude Fable 5.1 replaces Fable 5 as the default Fable model in Claude Code: 1M context window, $10/$50 per Mtok for input/output, and cache reads cut from $1 to $0.25 per Mtok. For workloads that lean on cached context, that's a 75% drop on the read side alone.
featurehigh
Added Claude Fable 5.1 model with 1M context as new default Fable model
Users now have access to Claude Fable 5.1 (claude-fable-5-1) as the default Fable model with 1M context window, priced at $10/$50 per Mtok with $0.25/Mtok cache reads for improved performance.
Evidence
+Added Claude Fable 5.1 (<code>claude-fable-5-1</code>), now the default Fable model — 1M context, $10/$50 per Mtok with $0.25/Mtok cache reads
Braintrust cut the monthly included credits on its main plan from $249 to $100, a 60% reduction. Heavy evaluation users will burn through the included allowance faster and need to either buy more credits or scale back eval volume.
pricinghigh
Monthly included credits reduced from $249 to $100
Paying users on Braintrust's main plan will receive fewer included credits per month, requiring either more frequent purchases or reduced evaluation volume.
Langfuse cut the default JWT session max age to 14 days, tagged as a breaking security fix. Sessions that used to stay alive longer now force a re-login after two weeks; automation that depends on long-lived sessions should budget for more frequent re-authentication.
otherhigh
Security fix: JWT session max age reduced to 14 days
All users will have sessions automatically expire after 14 days instead of previous default, requiring re-authentication for longer-lived sessions.
Evidence
+security(auth)!: reduce default JWT session max age to 14 days
Moonshot's V1 model series is off the Kimi API platform as of August 31; the pricing page no longer lists it at all. Anyone still pointed at a V1 model name needs to move to a current one now, without a grace period.
deprecationhigh
Moonshot V1 model series discontinued with platform sunset
Developers using Moonshot V1 will need to migrate to newer models by August 31 as the entire V1 platform is being sunset.
Evidence
−Moonshot V1
Classic generation model series; full platform sunset expected on August 31
Pinecone dropped the one-time $250 (1TB) bulk import credit for new Standard and Enterprise subscriptions; imports from object storage now bill at the standard $0.25/GB rate from the first byte. If you were budgeting a large first import around that credit, it's gone.
pricinghigh
Bulk import credit of $250 (1 TB) no longer offered for new subscriptions
New Standard and Enterprise customers will no longer receive the one-time $250 bulk import credit and will be billed at $0.25/GB for all imports from object storage.
Evidence
+Bulk import credit is no longer offered
New Standard and Enterprise subscriptions no longer receive the one-time $250 bulk import credit (1 TB). Imports from object storage are billed at the standard $0.25/GB rate.
The Kimi API now speaks the Anthropic API format alongside OpenAI's, so it can sit behind Claude Code and other Anthropic-SDK integrations without a translation shim in between.
apihigh
Kimi API now supports Anthropic API format in addition to OpenAI format
Developers can now use Kimi with the Anthropic SDK and Claude Code integrations, expanding compatibility beyond OpenAI ecosystem
Evidence
−Kimi API lets you interact with Kimi models and is compatible with the OpenAI API format.
+Kimi API lets you interact with Kimi models and is compatible with both the OpenAI and Anthropic API formats.
Pinecone's full-text search (BM25) and the new Documents API reach general availability on API version 2026-07: you can now combine full-text, dense-vector, and sparse-vector ranking in one schema-based index, plus bulk document operations.
featurehigh
Full-text search and Documents API reach general availability
Developers can now use production-ready full-text search (BM25) combined with dense and sparse vector search in single indexes, with advanced querying, filtering, and bulk document operations.
Evidence
+General availability: Full-text search and the Documents API
Full-text search is now generally available on API version 2026-07, built on the new Documents API.Create schema-based document indexes that combine full-text (BM25), dense-vector, and sparse-vector ranking fields in a single index.
If you self-host Zep's Neo4j MCP server with a custom NEO4J_DATABASE (anything other than the default 'neo4j'), earlier versions silently split your data: writes went to the configured database while reads and searches hit the server's home database. The fix makes every operation respect the configured database consistently, but check your home database for orphaned writes before assuming the upgrade alone fixes your data.
deprecationhigh
Neo4j MCP server behavior change: NEO4J_DATABASE setting now honored for all operations
Self-hosted Neo4j Enterprise deployments with custom database settings (non-default 'neo4j') may have experienced data split across databases. Users must explicitly verify NEO4J_DATABASE configuration and check for data that may have been written to the home database before upgrading.
Evidence
+WHO IS AFFECTED: Self-hosted Neo4j Enterprise deployments where NEO4J_DATABASE (or database: in config.yaml) was set to something other than neo4j. Standard setups using the default neo4j database need no action.
Before: The MCP server ignored your database setting; add_memory / search could hit the server's home database while other operations used the configured one — data could be split across two databases.
After: All operations consistently use the configured database. Set NEO4J_DATABASE explicitly in .env or config.yaml:
NEO4J_DATABASE=your_database_name
Migration: If you previously ran with a custom database name configured but data appeared missing, check your Neo4j home database — earlier writes may have landed there and may need to be migrated.
Cursor's Cloud Agents can now run on Vercel Sandbox: scale-to-zero workers, isolated microVMs, and durable orchestration, all on infrastructure you control instead of Cursor's own hosting.
featurehigh
Cursor Cloud Agents now supported on Vercel Sandbox infrastructure
Users can run Cursor Cloud Agents on Vercel's managed infrastructure with scale-to-zero workers and isolated microVMs, enabling more flexible AI agent deployment options.
Evidence
+Cursor Cloud Agents can now run in Vercel Sandbox
Run Cursor Cloud Agents on infrastructure you control with Vercel Sandbox, using scale-to-zero workers, isolated microVMs, and durable orchestration.
Pinecone's cohere-rerank-3.5 auto-migration, originally set for July 31, actually landed a month later on August 31: requests are now silently served by cohere-rerank-4-fast if you never moved off it yourself. If your rerank results shifted in the past week, this is why.
deprecationhigh
cohere-rerank-3.5 deprecation deadline moved from July 31 to August 31, 2026
Customers have one additional month (extended from July 31 to August 31) to migrate rerank requests to cohere-rerank-4-fast before automatic migration occurs.
Evidence
−Through July 31, 2026: cohere-rerank-3.5 continues to serve requests as before.
August 1, 2026: requests to cohere-rerank-3.5 are automatically served by cohere-rerank-4-fast.
+Through August 30, 2026: cohere-rerank-3.5 continues to serve requests as before.
August 31, 2026: requests to cohere-rerank-3.5 are automatically served by cohere-rerank-4-fast.
Neon added an AWS Europe (Frankfurt, eu-central-1) region, alongside the existing US East (Ohio). European teams now have a local option for latency and data-residency requirements.
featurehigh
Neon backend now available in AWS Europe (Frankfurt)
European users can now host Neon backends in their region (AWS eu-central-1), reducing latency and improving compliance with data residency requirements.
Evidence
+The Neon backend is now available in AWS Europe (Frankfurt) (aws eu central 1), alongside AWS US East (Ohio)