GitHub transforms Copilot into a multi-agent platform with Claude and Codex, Anthropic takes a stand against advertising in AI assistants, Apple integrates Claude Agent SDK into Xcode 26.3, Mistral launches two transcription models, and ElevenLabs triples its valuation.
GitHub Copilot: Claude and Codex in Agent HQ
February 4 โ GitHub opens Agent HQ to third-party agents: Claude (Anthropic) and OpenAI Codex join Copilot in public preview. Developers can now launch coding sessions with multiple agents from a unified interface.
| Feature | Description |
|---|---|
| Multi-agent | Copilot, Claude, and Codex in the same workspace |
| Surfaces | github.com, GitHub Mobile, VS Code (1.109+) |
| Assignment | Assign an agent to an issue or PR, or all three to compare |
| Async sessions | Each session consumes 1 premium request, results as draft PR |
| Mentions | @Copilot, @Claude, @Codex in comments |
Availability: Copilot Pro+ and Copilot Enterprise. Agents from Google, Cognition, and xAI are announced coming soon.
VS Code 1.109: multi-agent locally
On the same day, GitHub releases VS Code Copilot 1.109 with multi-agent support. Developers can launch local or asynchronous sessions on GitHub directly from the editor.
Codex in Agent HQ
On the OpenAI side, the team details the architecture of the Codex App Server, the bidirectional JSON-RPC server powering the Codex app, CLI, IDE extension, and web. It is this same infrastructure that allows Codex to function as an agent in Agent HQ.
๐ GitHub Blog | Changelog | VS Code 1.109 | Codex App Server
Anthropic: Claude ad-free and Xcode SDK
Claude will remain ad-free
February 4 โ Anthropic publishes a detailed essay explaining why Claude will never integrate advertising. The argument is structural: a model funded by ads would have incentives incompatible with authentic assistance.
| Argument | Detail |
|---|---|
| User trust | AI conversations are more intimate than a web search |
| Incentives | Ads optimize for engagement, not real utility |
| Expansion | Advertising in products tends to expand over time |
| Business model | Revenue via subscriptions and enterprise contracts |
Anthropic states that it wishes to develop agentic commerce (Claude acts for the user during purchases) and third-party integrations (Figma, Asana, Canva), but always at the userโs initiative, never an advertiserโs.
๐ Claude is a space to think
Apple Xcode 26.3: Native Claude Agent SDK
February 3 โ Apple integrates the Claude Agent SDK into Xcode 26.3, giving Apple developers the power of Claude Code directly within the IDE.
| Feature | Description |
|---|---|
| Visual verification | Claude captures Xcode Previews to iterate on SwiftUI interfaces |
| Multi-project reasoning | Complete exploration of project structure and Apple frameworks |
| Autonomous execution | Claude breaks down a goal into subtasks and iterates until completion |
| MCP Interface | Access to Xcode capabilities from Claude Code via the Model Context Protocol |
Availability: Xcode 26.3 Release Candidate for Apple Developer Program members.
Mistral: Voxtral Transcribe 2
February 4 โ Mistral launches Voxtral Transcribe 2, a family of two speech transcription models with diarization and ultra-low latency.
| Model | Usage | Price | Feature |
|---|---|---|---|
| Voxtral Mini Transcribe V2 | Batch | $0.003/min | SOTA (best score), diarization, 13 languages |
| Voxtral Realtime | Live | $0.006/min | Configurable sub-200ms latency, Apache 2.0 |
Voxtral Realtime
Native streaming architecture that transcribes audio on the fly. At 2.4 seconds delay, it matches the batch model. At 480ms, it remains within 1-2% error rate difference. The model is 4B parameters and runs on edge devices. Weights available under Apache 2.0 on Hugging Face.
Voxtral Mini Transcribe V2
At ~4% error rate on FLEURS, it is the best value on the market. It surpasses GPT-4o mini Transcribe, Gemini 2.5 Flash, Assembly Universal, and Deepgram Nova in accuracy, and processes audio ~3x faster than ElevenLabs Scribe v2 for a fifth of the price.
| Feature | Detail |
|---|---|
| Diarization | Speaker labels with precise timestamps |
| Context biasing | Up to 100 words/phrases to guide transcription |
| Word timestamps | Start/end per word for subtitles and audio search |
| Long audio | Up to 3 hours in a single request |
| 13 languages | EN, ZH, HI, ES, AR, FR, PT, RU, DE, JA, KO, IT, NL |
๐ Mistral Blog
ElevenLabs: 11B valuation
February 4 โ ElevenLabs raises 11 billion โ a tripling in one year.
| Detail | Value |
|---|---|
| Amount raised | $500M |
| Valuation | $11B |
| Lead investor | Sequoia Capital |
| Total raised | $781M since 2022 |
| Projected ARR | $330M+ end of 2025 |
Among investors: Lightspeed Venture Partners, BOND, Evantic Capital (new), and Andreessen Horowitz and ICONIQ significantly increasing their stake.
Funds will be dedicated to ElevenAgents, the enterprise platform for voice and conversational agents, as well as research on emotional models, dubbing, and audio general intelligence.
๐ ElevenLabs Blog
Manus Academy: Product Management courses
February 5 โ Manus launches Manus Academy, a certification training platform to learn how to use Manus in professional contexts: Business Analysis, Product Management, and Finance/Private Equity.
๐ Manus Academy
What this means
GitHub takes a decisive step by transforming Copilot into an open multi-agent platform. Being able to launch Claude, Codex, and Copilot on the same issue and compare results is a paradigm shift for developers: the coding agent becomes a choice, not a monopoly.
Anthropic clearly positions itself on values with the refusal of advertising, while Apple validates the Claude Agent SDK as a standard for iOS/macOS development. Mistral hits hard on transcription with a competitive open-weight model, and ElevenLabs confirms its status as the AI audio leader with a valuation tripling in one year.