All 5 AI packs for $39 →
← All posts
News · Coding · Agents

This week in AI tools: GitHub Copilot Fusion, Claude API changes, Replit Cloud Agents and Runway Dev plugins

Published · Facts checked against official vendor pages on November 3, 2026 · Rankings are independent of sponsorships

The biggest news since our last roundup came from GitHub Universe on October 28, but the useful changes were spread across the tools developers actually use: Copilot now routes between models for you, Claude's API got cheaper-to-use conversation handling and a real cost API, and several agents now work against your existing codebase instead of a blank project. Everything below comes from the vendors' own announcements, changelogs and docs, with dates.

Tools in this post

AGitHub CopilotSClaudeSClaude CodeAReplitBLovableBBolt.newARunwaySSunoSElevenLabsSChatGPTAGeminiBKimi (Moonshot AI)SCursor

The short version

GitHub Copilot Fusion (Oct 28) is generally available: pick Fusion in the model picker and Copilot routes and orchestrates models for each task, at Standard, High or Max effort.

GitHub Copilot app and Agent Merge (Oct 28): multi-repo Projects and desktop computer use in public preview, and Agent Merge now fixes failing checks and merge conflicts on pull requests.

Claude API (Oct 28 to Nov 2): message threads in beta, 128 MB requests, a new Usage and Cost Analytics API in beta, and Opus 4.8 fast mode retiring on November 30.

Replit Cloud Agents (Oct 27) turn a plain-language request into a tested pull request on your existing repository, for Pro and Enterprise.

Runway Dev plugins (Nov 2) put Runway's media API inside Claude Code, Codex and Cursor; Suno MCP (Oct 28) and ElevenLabs SFX v3 (Nov 2) bring audio generation into agent workflows.

GitHub Copilot Fusion: stop picking models by hand

With Fusion, you are not choosing a model but a system. For each prompt it decides which models to use and how they work together: one model alone, a cascade where an efficient model drafts and a quality check decides whether to escalate, or a critique loop where a second model reviews the first model's draft. You only choose the effort level: Standard (everyday coding), High (harder tasks) or Max (the most demanding work).

GitHub's headline number is that Fusion Max scored 60 on the Artificial Analysis Coding Agent Index, within two points of Claude Code running Opus 5.5 at Max, at about half the average API cost per task, about half as many tokens and about 40% less time. That is GitHub's report of a third-party index, so treat it as a claim to test on your own repos. On cost, you are billed for every model Fusion uses at that model's standard rate, with a limited-time 30% discount on model costs for paid Copilot plans. It works on all Copilot plans in VS Code, the Copilot app, Copilot CLI and JetBrains.

If you manage a team, administrators can make Fusion the default model through Enterprise managed settings. Full plan prices are on our GitHub Copilot page; this week's change to enterprise billing is in the Nov 2 pricing roundup.

Copilot app Projects, computer use and Agent Merge

Projects in the GitHub Copilot app are in public preview. You give Copilot a goal, add context (repositories, design mocks, Slack threads, meeting notes), pick models and tools, and it runs as many agents as the work needs, returning pull requests for review. Computer use, in public preview on macOS and Windows, lets Copilot click through desktop apps that have no API; it asks for approval first and organizations can switch it off.

The change most people will notice first is Agent Merge on pull requests. Select Watch and fix in the merge box and Copilot works in a cloud environment to fix failing Actions workflows, merge conflicts or both, and it can be paired with auto-merge so the PR merges once requirements are met. It needs a paid Copilot plan and respects organization policies. GitHub also said Copilot SDK is becoming the Microsoft Copilot SDK, with no immediate changes for existing users. Our free copilot-instructions-tests file is a good way to give those agents your testing rules.

Claude API: threads, bigger requests and a cost API

Anthropic's Claude Platform release notes logged four developer changes this week. Message threads (Nov 2, beta) let the API keep a conversation on the server so each later request sends only the new messages, which Anthropic says gives smaller requests and a lower time to first token for long conversations and agent loops. A thread expires 5 minutes after its last request and is deleted within an hour of last use.

Request size went from 32 MB to 128 MB for the Messages and Token Counting APIs (Oct 28), which matters for image-heavy and screenshot-heavy agent turns. The Usage and Cost Analytics API (Oct 29, beta) reports tokens and dollars for Console and Enterprise organizations, grouped by model, workspace, API key or the user behind a request; the older usage and cost endpoints are now legacy but still supported.

The one with a deadline: fast mode for Claude Opus 4.8 is deprecated and will be removed on November 30, 2026. After that, requests to claude-opus-4-8 with speed set to fast return an error, and Anthropic recommends moving to Claude Opus 5.5. If you use fast mode in Claude Code or through the API, check which model it is pinned to. See Claude for plans, and our claude-skill-ship-check Agent Skill, which reviews your live git diff before you commit.

Agents that work on your existing code

Replit Cloud Agents for existing repositories are now available for all Pro and Enterprise customers. Anyone on the team can describe a change, watch a live preview, and have the agent open a pull request on GitHub or a merge request on GitLab that follows the team's guidelines. Before anyone reviews it, the agent tests the change in a cloud sandbox and attaches screenshots and video, and it applies review feedback itself. The Slack integration (mention @Replit) is open to all customers, and Replit says a workspace-memory feature that proposes skill updates for engineers to approve is in early access. See Replit.

Lovable went the other direction and made it easier to arrive: you can import a project from Replit, Bolt, Base44, v0 or Manus, or an existing website, with the first import free and later ones counted as normal build usage. It also added persistent project and personal memory (Oct 30) and a downloadable SBOM (Nov 1). See Lovable and Bolt.

Runway, Suno and ElevenLabs reach into agent workflows

Runway Dev plugins (Nov 2) for Claude Code, Codex and Cursor connect your agent to your Runway Dev account over MCP, with skills for the official SDKs, Model Routers, Characters and Recipes. Sign-in happens through your account, and Runway says no API key is stored in the plugin. See Runway.

Suno MCP (beta, Oct 28) connects Suno to Claude, ChatGPT, Claude Code and other MCP clients so you can write songs, turn a voice memo into a track, or make covers and mashups from chat; songs use your credits and access is by early-access sign-up. ElevenLabs SFX v3 (Nov 2) is a new sound-effects model, selected with model_id eleven_text_to_sound_v3, that follows multi-part prompts more closely than v2; looping still needs v2. See Suno and ElevenLabs.

Smaller updates worth knowing

A few more items from official changelogs this week:

  • OpenAI released GPT-6 Rosalind (Oct 30), a life-sciences reasoning model available only to organizations approved through its trusted-access program, so it is not a general API release. See ChatGPT.
  • Google's Gemini app can now generate, preview and edit documents, PDFs, presentations, spreadsheets and code inside the chat (Oct 29). PDFs work on web and mobile; the other formats are web-only. See Gemini.

Follow-ups from last week

Kimi K3.1 Pro weights: Moonshot said full weights would arrive by November 4. As of this morning the moonshotai page on Hugging Face still lists Kimi K3 (updated Sep 1) as its newest large model, so we are waiting one more day before calling it. See Kimi.

GPT-6.1 Terra in Cursor: Cursor's models page, checked today, still lists GPT-5.6 Terra, Sol and Luna and does not list GPT-6.1 Terra, though GitHub Copilot has offered it since October 26. See Cursor.

What this means for our rankings

No tier changes this week. Fusion, Agent Merge and Cloud Agents are meaningful additions, but we move a tool's tier only after a human review, not on announcement day. Every item above has a dated, sourced entry in the changelog. If you are building with agents, our Agent Builder pack and Claude Code Agents pack cover the configs these tools read, and the free AGENTS.md and CLAUDE.md templates are a quick start.

FAQ

What is GitHub Copilot Fusion?
Copilot Fusion is a model-picker option, generally available since October 28, 2026, where Copilot routes and orchestrates one or more models for each task instead of you choosing one. You pick an effort level: Standard, High or Max. You are billed for each model it uses at that model's standard rate, with a limited-time 30% discount on model costs for paid Copilot plans.
When is Claude Opus 4.8 fast mode going away?
Anthropic deprecated fast mode for Claude Opus 4.8 on October 30, 2026 and will remove it on November 30, 2026. After that, requests to claude-opus-4-8 with speed set to fast return an error. Anthropic recommends migrating to Claude Opus 5.5.
What does Agent Merge do?
Agent Merge watches a pull request and, in a cloud development environment, fixes blockers such as failing Actions workflows and merge conflicts. It is now built into GitHub pull requests for users on a paid Copilot plan, subject to organization and enterprise policies, and can be combined with auto-merge.
Who can use Replit Cloud Agents on an existing repository?
Replit says Cloud Agents for existing repositories are available to all Pro and Enterprise customers, while the Slack integration is available to all customers. Agents open pull requests on GitHub or merge requests on GitLab with tested changes and screenshots or video.

Free and paid rules for your agents

Related changelog

Sources

Prices change often. Always confirm on the vendor's site before buying. See the full tier lists and the changelog.

+ Your logo here