v2.1.163: Versiones administradas, /plugin list y feedback de hooks sin error
Las nuevas configuraciones administradas requiredMinimumVersion y requiredMaximumVersion hacen que Claude Code rechace el inicio si la versión instalada queda fuera del rango aprobado — ideal para rollouts empresariales controlados. El nuevo /plugin list admite filtros --enabled y --disabled. El atajo “c to copy” en /btw copia la respuesta en markdown al portapapeles. Los hooks Stop y SubagentStop ahora pueden devolver hookSpecificOutput.additionalContext para dar feedback a Claude sin que el turno se etiquete como error de hook.
v2.1.162: Monitoreo de agentes con waitingFor en JSON y pill de Remote Control persistente
claude agents --json now includes a waitingFor field showing exactly what is blocking a waiting session — useful for debugging multi-agent pipelines programmatically. The Remote Control feature now renders as a persistent footer pill instead of a transient startup message. /effort confirms the chosen level and persists it as the new default.
AgentKit GA: Agent Builder visual, Connector Registry y ChatKit embebible
OpenAI launched AgentKit — a complete toolkit for building, deploying, and optimizing multi-agent applications. Ships three components: Agent Builder (a visual canvas for designing multi-agent workflows), Connector Registry (centralized management of tool connections), and ChatKit (an embeddable chat component for surfacing agent-powered experiences in any product). Targets developers who want to build on top of the same agent runtime that powers Codex.
GPT-4.5 se retira el 27 jun — 22 días para migrar a GPT-5.x
GPT-4.5 will be retired from ChatGPT on June 27, 2026. API users should migrate to a GPT-5.x model before the cutoff: GPT-5-Codex for agentic coding tasks, GPT-5.5 for general-purpose high-capability work, or GPT-5-Rosalind for life sciences. GPT-4.5 was OpenAI's last “traditional” non-reasoning model before the o-series and 5.x generation.
⚠️ Quedan 13 días: el 18 jun se corta el servicio para cuentas individuales — migrar a Antigravity
Gemini Code Assist IDE Extensions and Gemini CLI stop serving requests on June 18, 2026 for Individual, Google AI Pro, and Google AI Ultra tiers. Migrate to Antigravity and the Antigravity CLI before that date. Standard and Enterprise tiers are unaffected. Antigravity offers MCP support, multi-agent orchestration, subagents, hooks, and scheduled background tasks — comparable feature depth to Claude Code.
Gemini 3.5 Pro: GA esperada este mes — 2M tokens, razonamiento Deep Think
Gemini 3.5 Pro remains in limited Vertex AI preview with no confirmed GA date. The model targets a 2M-token context window and Deep Think reasoning mode — covering the use cases previously handled by Gemini Ultra. Sundar Pichai committed to a June 2026 GA at Google I/O; no formal launch announcement has been made yet.
Ventana de 1M tokens y razonamiento configurable disponibles en VS Code, CLI y app
GitHub Copilot now supports a one-million-token context window — enabling work across large codebases, long documents, and complex multi-file projects without losing context. Configurable reasoning levels let you dial the balance between speed and depth, with extended thinking available for architectural design and complex debugging. Both features are live in VS Code, Copilot CLI, and the Copilot desktop app. Note: higher context and reasoning levels consume more GitHub AI Credits per interaction.
Code Review: Agent Skills con MCP y nuevo nivel de análisis medio para PRs complejos
Two public preview features expand Copilot code review. Agent Skills + MCP support let teams create custom review skills that invoke internal tools and pull context from issue trackers, documentation, and any MCP server — surfacing knowledge that lives outside the diff. The new medium analysis tier routes complex pull requests to a higher-reasoning model alongside the existing low (fast/cost-efficient) tier. Available for Copilot Pro+, Business, and Enterprise.
MAI-Code-1: el primer modelo de codificación propio de Microsoft, ya activo en Copilot y VS Code
MAI-Code-1 is Microsoft's first internally-built coding model, debuted at Build 2026 and already live in GitHub Copilot and VS Code for code completions. This marks Microsoft's transition from being purely a model reseller to competing at the coding-model layer — part of the broader MAI (Microsoft AI) model family trained entirely in-house. No benchmark disclosures yet; Microsoft said more details are coming at //Build sessions.
Microsoft IQ: capa de conocimiento empresarial GA en Copilot, Foundry y Copilot Studio
Microsoft IQ is a new context layer that grounds AI agents in enterprise data: Work IQ (Microsoft 365 signals), Fabric IQ (structured business data from Fabric), and Web IQ (real-time web grounding). Now generally available across GitHub Copilot, Microsoft Foundry, and Copilot Studio — enabling agents to answer grounded in actual workplace knowledge, not just pre-training data.
Windsurf se rebautiza como Devin Desktop con ACP abierto y Devin Local en Rust (30% más eficiente)
Windsurf officially rebranded to Devin Desktop on June 2, 2026. Headline change: the Agent Client Protocol (ACP) — an open agent-editor communication standard already adopted by JetBrains, Google, GitHub, and 25+ agent tools on day one. The Cascade engine was rewritten in Rust as Devin Local (30% more token-efficient, supports subagents). Cascade deprecated July 1. Plans and pricing unchanged.
Línea de Tiempo — Lanzamientos de Modelos
| Fecha | Modelo | Proveedor | Tier | Tipo | Notas |
|---|---|---|---|---|---|
| 2026-06-01 | Qwen3.7-Plus | Alibaba / Qwen | A+ | Propietario | Agente multimodal GA. GUI grounding ScreenSpot Pro 79.0. Workflows GUI+CLI híbridos. $2.50/$7.50 por 1M tokens. |
| 2026-05-28 | Claude Opus 4.8 | Anthropic | S+ | Propietario | Modelo GA más capaz. 1M tokens. Intelligence Index 61.4 (#1 global). Fast mode 2.5× velocidad, 3× más barato. |
| 2026-05-19 | Gemini 3.5 Flash | S | Propietario | GA en Google I/O 2026. Inteligencia frontier a 4× la velocidad. 76.2% Terminal-Bench 2.1. $1.50/$9 por 1M tokens. | |
| 2026-05-20 | Qwen3.7 Max | Alibaba / Qwen | A+ | Propietario | Flagship texto. Intelligence Index 56.6 (#1 en China). 1M tokens. 60.6% SWE-Bench Pro. Menor tasa alucinaciones: 22.9%. |
| 2026-04-20 | Kimi K2.6 | Moonshot AI | A+ | Abierto | 1T params (32B activos), open-weight. 58.6% SWE-Bench Pro. Swarm: 300 sub-agentes, 4.000 pasos coordinados. |
| 2026-04-24 | DeepSeek V4-Pro | DeepSeek | S | Abierto | MoE 1.6T params, 32T tokens, contexto 1M nativo. Variante V4-Flash para alta velocidad. |
| 2026-06-30 | Gemini 3.5 Pro | S+ | Próximo | GA esperada junio 2026. Objetivo: 2M tokens, razonamiento Deep Think. Preview limitado en Vertex. |
Tendencias Destacadas
Microsoft Build 2026 marcó un punto de inflexión: con MAI-Code-1, Microsoft dejó de ser solo distribuidor de modelos de terceros y comenzó a competir en la capa de modelos de codificación. El lanzamiento simultáneo de Microsoft IQ — que ancla agentes en datos empresariales reales desde M365, Fabric y la web — posiciona a GitHub/Copilot como una plataforma de inteligencia empresarial, no solo un asistente de código.
La carrera de ventana de contexto se intensificó: GitHub Copilot alcanza 1M tokens con razonamiento configurable, uniéndose a Claude Opus 4.8 y Gemini 3.5 Flash. El patrón emergente es contexto grande por defecto más profundidad de razonamiento como control del desarrollador — gestionando el costo de AI Credits según la complejidad de cada tarea.
El rebrand de Windsurf a Devin Desktop y el lanzamiento de ACP como estándar abierto señalan la maduración del ecosistema de agentes de codificación: los agentes comienzan a estandarizar sus protocolos de comunicación con editores, del mismo modo que MCP estandarizó la integración de herramientas. La adopción de ACP en el día del lanzamiento por JetBrains, Google y GitHub sugiere que el estándar tiene tracción real.
v2.1.163: Managed version pinning, /plugin list, and hook feedback without error labels
New managed settings requiredMinimumVersion and requiredMaximumVersion cause Claude Code to refuse startup when the installed version falls outside the approved range — useful for controlled enterprise rollouts. New /plugin list with --enabled / --disabled filters. The “c to copy” shortcut in /btw copies raw markdown to the clipboard preserving formatting. Stop and SubagentStop hooks can now return hookSpecificOutput.additionalContext to give Claude feedback without the turn being labeled a hook error.
v2.1.162: Agent monitoring with waitingFor in JSON output and persistent Remote Control pill
claude agents --json now includes a waitingFor field showing exactly what is blocking a waiting session — useful for debugging multi-agent pipelines programmatically. The Remote Control feature now renders as a persistent footer pill instead of a transient startup message. /effort confirms the chosen level and persists it as the new default.
AgentKit GA: visual Agent Builder, Connector Registry, and embeddable ChatKit
OpenAI launched AgentKit — a complete toolkit for building, deploying, and optimizing multi-agent applications. Ships three components: Agent Builder (a visual canvas for designing multi-agent workflows), Connector Registry (centralized management of tool connections), and ChatKit (an embeddable chat component for surfacing agent-powered experiences in any product). Targets developers who want to build on top of the same agent runtime that powers Codex.
GPT-4.5 retires June 27 — 22 days to migrate to GPT-5.x
GPT-4.5 will be retired from ChatGPT on June 27, 2026. API users should migrate to a GPT-5.x model before the cutoff: GPT-5-Codex for agentic coding tasks, GPT-5.5 for general-purpose high-capability work, or GPT-5-Rosalind for life sciences. GPT-4.5 was OpenAI's last “traditional” non-reasoning model before the o-series and 5.x generation.
⚠️ 13 days left: Jun 18 end-of-service for individual accounts — migrate to Antigravity now
Gemini Code Assist IDE Extensions and Gemini CLI stop serving requests on June 18, 2026 for Individual, Google AI Pro, and Google AI Ultra tiers. Migrate to Antigravity and the Antigravity CLI before that date. Standard and Enterprise tiers are unaffected. Antigravity offers MCP support, multi-agent orchestration, subagents, hooks, and scheduled background tasks — comparable feature depth to Claude Code.
Gemini 3.5 Pro: GA expected this month — 2M-token context, Deep Think reasoning
Gemini 3.5 Pro remains in limited Vertex AI preview with no confirmed GA date. The model targets a 2M-token context window and Deep Think reasoning mode — covering the use cases previously handled by Gemini Ultra. Sundar Pichai committed to a June 2026 GA at Google I/O; no formal launch announcement has been made yet.
1M-token context window and configurable reasoning live in VS Code, CLI, and app
GitHub Copilot now supports a one-million-token context window — enabling work across large codebases, long documents, and complex multi-file projects without losing context. Configurable reasoning levels let you dial the balance between speed and depth, with extended thinking available for architectural design and complex debugging. Both features are live in VS Code, Copilot CLI, and the Copilot desktop app. Note: higher context and reasoning levels consume more GitHub AI Credits per interaction.
Code Review: Agent Skills with MCP and a new medium analysis tier for complex PRs
Two public preview features expand Copilot code review. Agent Skills + MCP support let teams create custom review skills that invoke internal tools and pull context from issue trackers, documentation, and any MCP server — surfacing knowledge that lives outside the diff. The new medium analysis tier routes complex pull requests to a higher-reasoning model alongside the existing low (fast/cost-efficient) tier. Available for Copilot Pro+, Business, and Enterprise.
MAI-Code-1: Microsoft's first in-house coding model, now live in Copilot and VS Code
MAI-Code-1 is Microsoft's first internally-built coding model, debuted at Build 2026 and already live in GitHub Copilot and VS Code for code completions. This marks Microsoft's transition from being purely a model reseller to competing at the coding-model layer — part of the broader MAI (Microsoft AI) model family trained entirely in-house. No benchmark disclosures yet; Microsoft said more details are coming at //Build sessions.
Microsoft IQ: enterprise knowledge layer GA across Copilot, Foundry, and Copilot Studio
Microsoft IQ is a new context layer that grounds AI agents in enterprise data: Work IQ (Microsoft 365 signals), Fabric IQ (structured business data from Fabric), and Web IQ (real-time web grounding). Now generally available across GitHub Copilot, Microsoft Foundry, and Copilot Studio — enabling agents to answer grounded in actual workplace knowledge, not just pre-training data.
Windsurf rebrands to Devin Desktop with open ACP and Devin Local in Rust (30% more token-efficient)
Windsurf officially rebranded to Devin Desktop on June 2, 2026. Headline change: the Agent Client Protocol (ACP) — an open agent-editor communication standard already adopted by JetBrains, Google, GitHub, and 25+ agent tools on day one. The Cascade engine was rewritten in Rust as Devin Local (30% more token-efficient, supports subagents). Cascade deprecated July 1. Plans and pricing unchanged.
Model Launches Timeline
| Date | Model | Provider | Tier | Type | Notes |
|---|---|---|---|---|---|
| 2026-06-01 | Qwen3.7-Plus | Alibaba / Qwen | A+ | Proprietary | GA multimodal GUI agent. ScreenSpot Pro 79.0, hybrid GUI+CLI agentic workflows. $2.50/$7.50 per 1M tokens. |
| 2026-05-28 | Claude Opus 4.8 | Anthropic | S+ | Proprietary | Most capable GA model. 1M-token context. Intelligence Index 61.4 (#1 globally). Fast mode at 2.5× speed, 3× cheaper. |
| 2026-05-19 | Gemini 3.5 Flash | S | Proprietary | GA at Google I/O 2026. Frontier intelligence at 4× the speed. 76.2% Terminal-Bench 2.1. $1.50/$9 per 1M tokens. | |
| 2026-05-20 | Qwen3.7 Max | Alibaba / Qwen | A+ | Proprietary | Text-only flagship. Intelligence Index 56.6 (#1 Chinese model). 1M tokens. 60.6% SWE-Bench Pro. Lowest hallucination rate: 22.9%. |
| 2026-04-20 | Kimi K2.6 | Moonshot AI | A+ | Open | 1T params (32B active), open-weight. 58.6% SWE-Bench Pro. Agent swarm: 300 sub-agents, 4,000 coordinated steps. |
| 2026-04-24 | DeepSeek V4-Pro | DeepSeek | S | Open | 1.6T-parameter MoE, 32T training tokens, native 1M-token context. V4-Flash variant for high-throughput. |
| 2026-06-30 | Gemini 3.5 Pro | S+ | Upcoming | GA expected June 2026. Targets 2M-token context, Deep Think reasoning. Limited Vertex preview. |
Notable Trends
Microsoft Build 2026 marked an inflection point: with MAI-Code-1, Microsoft stopped being purely a model reseller and began competing at the coding-model layer itself. The simultaneous launch of Microsoft IQ — grounding agents in real enterprise data from M365, Fabric, and the web — repositions GitHub/Copilot as an enterprise intelligence platform, not just a coding assistant.
The context-window arms race intensified: GitHub Copilot reached 1M tokens with configurable reasoning, joining Claude Opus 4.8 and Gemini 3.5 Flash at that scale. The emerging pattern is large context as the default plus configurable reasoning depth as a developer dial — letting teams manage AI Credit costs based on task complexity rather than paying the same rate for everything.
Windsurf's rebrand to Devin Desktop and the ACP open standard signal agent-ecosystem maturation: coding agents are starting to standardize their editor communication protocols, just as MCP standardized tool integration. ACP's day-one adoption by JetBrains, Google, and GitHub suggests genuine traction and the potential to become the interoperability layer the agentic coding ecosystem has been missing.
v2.1.163: Version administrée, /plugin list et feedback de hook sans erreur
Les nouveaux paramètres administrés requiredMinimumVersion et requiredMaximumVersion font refuser le démarrage si la version est hors de la plage approuvée. Nouveau /plugin list avec filtres --enabled / --disabled. Le raccourci “c to copy” dans /btw copie le markdown brut. Les hooks Stop et SubagentStop peuvent retourner hookSpecificOutput.additionalContext sans erreur de hook.
v2.1.162: Surveillance agents avec waitingFor en JSON et pill Remote Control persistant
claude agents --json now includes a waitingFor field showing exactly what is blocking a waiting session — useful for debugging multi-agent pipelines programmatically. The Remote Control feature now renders as a persistent footer pill instead of a transient startup message. /effort confirms the chosen level and persists it as the new default.
AgentKit GA: Agent Builder visuel, Connector Registry et ChatKit intégrable
OpenAI launched AgentKit — a complete toolkit for building, deploying, and optimizing multi-agent applications. Ships three components: Agent Builder (a visual canvas for designing multi-agent workflows), Connector Registry (centralized management of tool connections), and ChatKit (an embeddable chat component for surfacing agent-powered experiences in any product). Targets developers who want to build on top of the same agent runtime that powers Codex.
GPT-4.5 se retire le 27 juin — 22 jours pour migrer vers GPT-5.x
GPT-4.5 will be retired from ChatGPT on June 27, 2026. API users should migrate to a GPT-5.x model before the cutoff: GPT-5-Codex for agentic coding tasks, GPT-5.5 for general-purpose high-capability work, or GPT-5-Rosalind for life sciences. GPT-4.5 was OpenAI's last “traditional” non-reasoning model before the o-series and 5.x generation.
⚠️ 13 jours restants: coupure de service le 18 juin pour les comptes individuels — migrer vers Antigravity
Gemini Code Assist IDE Extensions and Gemini CLI stop serving requests on June 18, 2026 for Individual, Google AI Pro, and Google AI Ultra tiers. Migrate to Antigravity and the Antigravity CLI before that date. Standard and Enterprise tiers are unaffected. Antigravity offers MCP support, multi-agent orchestration, subagents, hooks, and scheduled background tasks — comparable feature depth to Claude Code.
Gemini 3.5 Pro: GA attendue ce mois — contexte 2M tokens, raisonnement Deep Think
Gemini 3.5 Pro remains in limited Vertex AI preview with no confirmed GA date. The model targets a 2M-token context window and Deep Think reasoning mode — covering the use cases previously handled by Gemini Ultra. Sundar Pichai committed to a June 2026 GA at Google I/O; no formal launch announcement has been made yet.
Fenêtre 1M tokens et raisonnement configurable disponibles dans VS Code, CLI et app
GitHub Copilot now supports a one-million-token context window — enabling work across large codebases, long documents, and complex multi-file projects without losing context. Configurable reasoning levels let you dial the balance between speed and depth, with extended thinking available for architectural design and complex debugging. Both features are live in VS Code, Copilot CLI, and the Copilot desktop app. Note: higher context and reasoning levels consume more GitHub AI Credits per interaction.
Code Review: Agent Skills avec MCP et nouveau niveau d'analyse moyen pour les PRs complexes
Two public preview features expand Copilot code review. Agent Skills + MCP support let teams create custom review skills that invoke internal tools and pull context from issue trackers, documentation, and any MCP server — surfacing knowledge that lives outside the diff. The new medium analysis tier routes complex pull requests to a higher-reasoning model alongside the existing low (fast/cost-efficient) tier. Available for Copilot Pro+, Business, and Enterprise.
MAI-Code-1: le premier modèle de codage maison de Microsoft, désormais actif dans Copilot et VS Code
MAI-Code-1 is Microsoft's first internally-built coding model, debuted at Build 2026 and already live in GitHub Copilot and VS Code for code completions. This marks Microsoft's transition from being purely a model reseller to competing at the coding-model layer — part of the broader MAI (Microsoft AI) model family trained entirely in-house. No benchmark disclosures yet; Microsoft said more details are coming at //Build sessions.
Microsoft IQ: couche de connaissance entreprise disponible dans Copilot, Foundry et Copilot Studio
Microsoft IQ is a new context layer that grounds AI agents in enterprise data: Work IQ (Microsoft 365 signals), Fabric IQ (structured business data from Fabric), and Web IQ (real-time web grounding). Now generally available across GitHub Copilot, Microsoft Foundry, and Copilot Studio — enabling agents to answer grounded in actual workplace knowledge, not just pre-training data.
Windsurf devient Devin Desktop avec ACP ouvert et Devin Local en Rust (30% plus efficient)
Windsurf officially rebranded to Devin Desktop on June 2, 2026. Headline change: the Agent Client Protocol (ACP) — an open agent-editor communication standard already adopted by JetBrains, Google, GitHub, and 25+ agent tools on day one. The Cascade engine was rewritten in Rust as Devin Local (30% more token-efficient, supports subagents). Cascade deprecated July 1. Plans and pricing unchanged.
Chronologie des Lancements de Modèles
| Date | Modèle | Fournisseur | Tier | Type | Notes |
|---|---|---|---|---|---|
| 2026-06-01 | Qwen3.7-Plus | Alibaba / Qwen | A+ | Propriétaire | Agent GUI multimodal GA. ScreenSpot Pro 79.0, workflows GUI+CLI hybrides. $2.50/$7.50 / 1M tokens. |
| 2026-05-28 | Claude Opus 4.8 | Anthropic | S+ | Propriétaire | Modèle GA le plus capable. 1M tokens. Intelligence Index 61.4 (#1 mondial). Fast mode 2.5× vitesse, 3× moins cher. |
| 2026-05-19 | Gemini 3.5 Flash | S | Propriétaire | GA à Google I/O 2026. Intelligence frontier 4× la vitesse. 76.2% Terminal-Bench 2.1. $1.50/$9 / 1M tokens. | |
| 2026-05-20 | Qwen3.7 Max | Alibaba / Qwen | A+ | Propriétaire | Flagship texte. Intelligence Index 56.6 (#1 chinois). 1M tokens. 60.6% SWE-Bench Pro. Taux d'hallucination le plus bas: 22.9%. |
| 2026-04-20 | Kimi K2.6 | Moonshot AI | A+ | Ouvert | 1T params (32B actifs), open-weight. 58.6% SWE-Bench Pro. Essaim: 300 sous-agents, 4 000 étapes coordonnées. |
| 2026-04-24 | DeepSeek V4-Pro | DeepSeek | S | Ouvert | MoE 1.6T paramètres, 32T tokens, contexte 1M natif. Variante V4-Flash haute performance. |
| 2026-06-30 | Gemini 3.5 Pro | S+ | À venir | GA attendue juin 2026. Cible: 2M tokens, raisonnement Deep Think. Preview limité Vertex. |
Tendances Notables
Microsoft Build 2026 a marqué un point d'inflexion: avec MAI-Code-1, Microsoft a cessé d'être un revendeur de modèles pour concurrencer au niveau des modèles de codage. Le lancement simultané de Microsoft IQ positionne GitHub/Copilot comme une plateforme d'intelligence d'entreprise.
La course aux fenêtres de contexte s'est intensifiée: GitHub Copilot atteint 1M tokens avec raisonnement configurable, rejoignant Claude Opus 4.8 et Gemini 3.5 Flash. Le modèle émergent: grand contexte par défaut avec profondeur de raisonnement ajustable par le développeur.
Le rebrand de Windsurf vers Devin Desktop et le lancement d'ACP signalent la maturation de l'écosystème: les agents de codage standardisent leurs protocoles de communication avec les éditeurs, comme MCP a standardisé l'intégration d'outils. L'adoption d'ACP dès le premier jour par JetBrains, Google et GitHub indique une vraie traction.