The Signal

One million tokens. Standard pricing. No catch.

On 13 March, Anthropic made the 1 million token context window generally available for Opus 4.6 and Sonnet 4.6. No special access, no pricing multiplier. A 900,000-token request costs the same per-token as a 9,000-token one.

This has been technically possible since Opus 4.6 launched in February, but only through the API with special access. Now it is the default across all interfaces. If you are on Max, Team, or Enterprise, you already have it. Pro users get 1M in Claude Code via /extra-usage. Full details on our Models page.

The media limit also jumped from 100 to 600 images or PDF pages per request. That is an entire due diligence data room in a single prompt.

What this means practically: The workflows that were limited by context — feeding an entire quarter of financials into one conversation, loading a complete agreement set with schedules and amendments, processing a full board pack with previous minutes and strategy documents — now work without workarounds. No more splitting documents across sessions. No more summarising chunks and losing nuance.

The honest caveat: More context does not automatically mean better output. Models can lose focus in very long documents — internal benchmarks show strong but imperfect retrieval at longer context lengths. For most operator tasks, well-structured prompts with the right 50,000 tokens will outperform dumping 500,000 tokens of everything. Use the expanded window for tasks that genuinely need comprehensive cross-referencing — contract review, due diligence, quarter-over-quarter financial analysis.

Also this week: Anthropic doubled usage limits through 28 March, launched a B2B marketplace, shipped interactive visualisations in chat, and announced a Sydney office. The pace is not slowing down. We cover each of these below.

Tool Review

Interactive Charts and Visualisations in Chat

In mid-March, Claude gained the ability to create interactive charts, diagrams, and data visualisations directly inline in conversations. Not in the Artifacts side panel. Not as static images. Inline, interactive, HTML/SVG-based elements that support hover and click.

This is available to all users including Free. It is in beta. Claude may generate visuals proactively when it thinks they would help, or you can request them explicitly.

How it differs from Artifacts

Artifacts remain permanent — you can download them, share them, iterate on them in the side panel. Interactive visuals are temporary. They appear inline in the conversation and evolve as the discussion continues. Think of them as Claude sketching on a whiteboard during a meeting, not producing a final deliverable.

What operators should use this for

Meeting prep. Ask Claude to visualise your pipeline data, revenue trends, or project timelines before a meeting. The interactive charts are sharp enough to screenshot for a slide, and the hover details give you talking points.

"Here is our Q1 revenue by region [paste data]. Create an interactive chart showing month-over-month trend with hover details for each data point. Highlight any region that declined."

Quick analysis. When you are reviewing data and want to see patterns before committing to a full dashboard. Claude builds the visualisation in seconds — faster than setting up a spreadsheet chart.

"Create a comparison chart of these three vendors' pricing tiers [paste data]. Make it interactive so I can hover to see feature differences at each tier."

Decision trees. For process documentation or team training. Claude generates clickable decision diagrams that are more useful than static flowcharts.

The honest take

This is genuinely useful for quick, disposable visualisations — the kind you need for five minutes in a meeting, not the kind you put in a board deck. For polished outputs, Artifacts are still the right tool. For "show me what this data looks like before I decide what to do with it," interactive visuals are faster than anything else.

The beta label is deserved. Complex charts can render inconsistently, and Claude sometimes generates visuals when text would be better. But for operators who regularly work with data, this is a meaningful quality-of-life improvement.

Operator Workflow

Making the Most of the March Usage Promotion

Anthropic is running a time-limited promotion from 13 March through 28 March: doubled usage limits during off-peak hours. This applies to Free, Pro, Max, and Team plans across Claude web, desktop, mobile, Cowork, Claude Code, Excel, and PowerPoint add-ins. Enterprise is excluded.

Off-peak hours: Outside 8am–2pm ET on weekdays. All day on weekends. No action required — it is automatic.

For our AEDT readers: The weekday off-peak window translates to roughly 11pm–5am AEDT. But weekends are doubled all day — and that is the real opportunity.

The timing is not coincidental — it follows a massive influx of new users during the Pentagon controversy. Whatever the motivation, the extra capacity is real.

A weekend batch processing workflow

Here is how to use the remaining weekends (22–23 March is your last full window) to get maximum value:

Step 1: Identify your backlog. What Cowork tasks have you been putting off because of rate limits? Document reviews, large data analyses, content generation runs — anything that chews through your usage allocation.

Step 2: Prepare your inputs. Gather the files, data, and prompts you need. Upload them to a Cowork task on Friday evening. The 1M context window means you can load substantially more material than before.

Step 3: Run heavy tasks on Saturday and Sunday. With doubled limits all day, you can run multiple substantial Cowork sessions without hitting your cap. Schedule your automated tasks to fire during these windows too.

Step 4: Use the results during the week. Your analyses, reports, and drafts are ready. Use your normal weekday allocation for lighter interactive work.

Tasks worth batching this weekend

  • Quarterly document review. Load an entire quarter's contracts, invoices, or compliance documents and have Claude analyse patterns.

  • Content calendar generation. Feed Claude your brand guidelines, past content, and competitor examples. Generate a month's worth of drafts.

  • Data analysis backlog. Those spreadsheets you have been meaning to analyse but keep deprioritising.

  • Knowledge base creation. Process your internal wikis, SOPs, and tribal knowledge into structured documentation.

Here is a prompt you can copy and paste for the quarterly document review:

"I've uploaded [X] documents from Q1 2026 — contracts, invoices, and compliance filings. Analyse the full set and produce: (1) a summary of key terms and obligations across all contracts, (2) any inconsistencies between documents, (3) items that require follow-up before end of quarter. Flag anything unusual. Format as a structured report with sections I can forward to my team."

The promotion ends 28 March. After that, limits return to normal.

Quick Hits

Claude Marketplace launched. On 16 March, Anthropic launched a B2B marketplace where enterprises can browse and purchase third-party tools built on Claude. Launch partners include Snowflake, GitLab, Harvey AI, Replit, Rogo, and Lovable. Anthropic is taking zero commission — positioning it as a distribution channel, not a revenue line. For operators, this is where vetted integrations will live going forward.

Anthropic opened a Sydney office. Their fourth in Asia-Pacific, joining Tokyo, Bengaluru, and Seoul. The executive team is visiting Australia later this month. For our readers: local presence typically means better regional support and, eventually, data residency options.

Claude Code reached v2.1.77. The past two weeks delivered ten releases. Highlights: /loop runs prompts on a recurring interval (e.g., /loop 5m check the deploy), /voice adds push-to-talk voice mode in 20 languages, /effort controls thinking depth (low/medium/high), and Opus 4.6 now defaults to medium effort with 64k token output. Opus 4 and 4.1 have been removed from the first-party API.

Memory landed on the Free plan. Chat Memory is now available to all users, plus an import tool that lets you bring saved memories from ChatGPT or other platforms. If you are helping a colleague evaluate Claude before committing to Pro, this removes a significant friction point.

Copilot Cowork is in research preview. The Microsoft product we flagged last issue is now in limited testing — Claude's Cowork technology running inside Outlook, Teams, and Excel at $30/user/month. Broader availability expected through the Frontier programme, with the full E7 bundle launching later. Still early, but if your organisation runs on M365, this is worth tracking.

Anthropic Institute launched. A new research body led by co-founder Jack Clark (new title: Head of Public Benefit), combining Anthropic's red team, societal impact, and economic research groups. Notable hires from Google DeepMind and UVA. Focus areas: AI's impact on jobs, security, and governance.

Claude Partner Network: $100M commitment. Launch partners include Accenture, Deloitte, Cognizant, and Infosys. Free membership. If your company works with any of these consulting firms, expect Claude integration proposals in your next engagement.

The Bottom Line

One thing to try this week: Use the doubled weekend limits to clear your analysis backlog.

Pick the task you have been putting off because it feels too big for a normal session — the quarterly review, the vendor comparison, the data cleanup. Load everything into a single Cowork session this Saturday (remember, you now have 1M context and doubled limits all weekend). Let Claude process it while you do something else.

The promotion ends 28 March. The 1M context window does not. But the combination of both, right now, is a window worth using.

Next week: What to expect from the Copilot Cowork rollout, and a deep dive into which plugin to install first (updated for the March changes).

Claude for Operators is an independent publication. Not affiliated with Anthropic.

Just switched to Claude? Start with our Operator's Toolkit — free guides for every workflow.

Got a workflow worth sharing? Reply to this email.

Keep Reading