✦ Claude Dispatch — Control Your Desktop Agent from Your Phone

Launched on 18 March 2026 as a research preview for Pro and Max plan subscribers, Claude Dispatch bridges the Claude mobile app and the Claude Desktop app into a single, persistent agent thread. Instead of each Cowork session starting fresh, Claude now retains context from previous tasks across devices — you can hand off work to your desktop, step away, and check in or redirect Claude from your phone whenever you like.

How Dispatch works

• Your phone runs the Claude mobile app — this is your remote control.
• Your desktop (awake, with Claude Desktop open) does the actual compute work: file edits, web searches, code execution.
• A single persistent thread connects both ends — Claude remembers the task it was working on when you send a follow-up message from your phone.
• You can send new instructions, ask for a status update, or course-correct mid-task without sitting at your desk.

Practical use cases

• Start a long research task before leaving the office, then check Claude's progress from the commute home.
• Trigger a code refactor in Cowork, go to a meeting, and confirm the result on mobile before merging.
• Queue up a batch of document summaries from your phone and find them ready when you sit down again.

✦ Claude Code Voice Mode — Push-to-Talk in the Terminal

Voice mode for Claude Code started rolling out on 3 March 2026 and is continuing its progressive rollout through mid-March. Available on Pro, Max, Team, and Enterprise plans, it lets you dictate prompts to Claude Code using a push-to-talk mechanism — no typing required. Speech is transcribed locally (no audio sent to Anthropic's servers), and the transcript appears in the terminal before Claude acts on it.

Activating voice mode

/voice          # toggle voice mode on/off in any Claude Code session

Once active, hold the spacebar to speak, release to send. Claude Code processes the transcript identically to a typed prompt — all the same tools, file access, and reasoning capabilities apply.

Key design decisions

• Push-to-talk, not always-on — no persistent microphone listening means no accidental commands and no privacy concerns about your terminal monitoring background conversation.
• Terminal-native — voice mode operates at the CLI layer, so it works regardless of which editor or IDE you have open behind it.
• Local STT — speech-to-text processing happens on-device; only the resulting text transcript is sent to Claude.
• Customisable keybinding — the default push-to-talk key is spacebar, but you can remap it in keybindings.json (key: voice:pushToTalk), e.g. to meta+k.

✦ Claude Partner Network — $100M Enterprise Adoption Push

On 12 March 2026, Anthropic launched the Claude Partner Network — a formalised channel programme backed by a $100 million commitment for 2026 (with more signalled for subsequent years). The programme connects large enterprises to a vetted network of consulting and professional services firms — including anchor partners Accenture, Deloitte, Cognizant, and Infosys — that specialise in deploying Claude at scale.

What the programme offers

• Training & enablement — joint go-to-market investment, sales enablement, and co-marketing support.
• Technical certification — the new Claude Certified Architect: Foundations credential is available immediately for solution architects building production Claude applications; additional certifications for sellers and developers are planned for later in 2026.
• Code Modernisation starter kit — a pre-built asset helping partners migrate legacy codebases and address technical debt for enterprise clients.
• Dedicated Applied AI engineers — Anthropic is scaling its partner-facing headcount fivefold, adding technical architects for complex deployments and localised go-to-market support across international markets.

Why it matters for developers

The partner programme accelerates enterprise Claude adoption, which in turn drives demand for developers with Claude API skills. The free-to-join certification track means individual developers can now earn a verifiable Anthropic credential — useful differentiation for consultants and agencies building Claude-powered products.

✦ Models API Upgrade — Query Capabilities Programmatically

The Claude Developer Platform's Models API received a significant upgrade in mid-March 2026: the GET /v1/models and GET /v1/models/{model_id} endpoints now return a richer response that includes max_input_tokens, max_tokens (max output), and a structured capabilities object for every available model. This removes the need to hard-code model limits in your application and makes it straightforward to build model-agnostic pipelines that adapt automatically as new models roll out.

New response fields

• max_input_tokens — the model's context window size (e.g. 1000000 for Opus 4.6 and Sonnet 4.6).
• max_tokens — maximum output tokens the model can generate (e.g. 128000 for Opus 4.6, 64000 for Sonnet 4.6).
• capabilities — a structured object listing supported features such as extended thinking, vision, tool use, streaming, and structured outputs.

Example — dynamic model selection

import anthropic

client = anthropic.Anthropic()

# List all available models with their capabilities
models = client.models.list()

for model in models.data:
    caps = model.capabilities
    if caps.get("extended_thinking") and model.max_input_tokens >= 1_000_000:
        print(f"{model.id}: 1M context + thinking ✓")

# Or fetch a single model's limits
m = client.models.retrieve("claude-sonnet-4-6")
print(f"Max input:  {m.max_input_tokens:,} tokens")
print(f"Max output: {m.max_tokens:,} tokens")

Best practices

• Don't hard-code limits — fetch them at startup or cache them with a short TTL so your app automatically picks up when a model's capabilities change.
• Gate features on capabilities — check capabilities.extended_thinking before setting thinking={"type": "adaptive"} to avoid errors on models that don't support it.
• Use max_input_tokens to chunk — if your input may exceed the context window, divide it by max_input_tokens to determine your chunk strategy dynamically.