Claude Review (2026)

What is Claude?

Claude is the assistant professionals reach for when the output actually matters. Anthropic's models — Claude Fable 5.1, released September 1, 2026, and now Claude Opus 5.5, the first of the Claude 5.5 family, shipped September 22 — have a reputation the benchmarks keep confirming: writing that sounds like a sharp human rather than a press release, and code that works on the first run more often than any competitor. It's why Claude has quietly become the enterprise favorite even while ChatGPT dominates consumer mindshare — though 'quietly' undersells it now: Anthropic's run-rate revenue crossed $47 billion in May 2026, announced alongside a Series H valuing the company just shy of a trillion dollars.

Opus 5.5 is the September story, and it matters more for subscribers than for benchmark-watchers. Anthropic's pitch is that it 'performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5' — the 40% is a workload-cost claim; the list price is $4/$20 per million tokens, 20% under Opus 5 and a fifth of Fable 5.1's $10/$50. The independent numbers came in above the pitch: Artificial Analysis scores it 58 on its Intelligence Index at max effort, first of 212 models and five clear of Fable 5.1 and GPT-6 Astra (both 53), and it took #1 on the Vals Index and on Cursor's CursorBench 4.0 (57.8% vs Fable 5.1's 51.8%). For a Pro subscriber the practical change is that Fable-class output is now the everyday Opus, and Anthropic raised five-hour limits on Pro and Max at the same time. Two footnotes for specialists: biology and life-sciences questions fall back to Opus 5 unless your organization is in the Life Sciences Verification Program, and most cybersecurity tasks route to Opus 4.8. Sonnet 5.5 and Haiku 5.5 'will follow in the coming weeks.'

The product is built around sustained work rather than one-off questions. Projects hold your documents and context permanently. Artifacts turn responses into live documents, code sandboxes, and small apps you can iterate on side-by-side with the conversation. The 200K-token-plus context window swallows entire codebases or a year of board minutes in one go, and Claude stays coherent across all of it — our long-document tests put it clearly ahead of ChatGPT. For developers, Claude Code extends the same models into a terminal agent that can plan and execute real engineering work. And Cowork — Claude's agentic workspace — reached mobile and web for all paid plans in August 2026 (still wearing a beta label): sessions run in Anthropic's cloud, follow you between devices, and scheduled tasks keep executing with your laptop closed. Late August pushed further: on desktop, Cowork now opens its own built-in browser — Claude navigates, clicks, and fills forms in a sandboxed side panel that doesn't touch your tabs or logins, and only sees saved passwords you choose to import — Claude in Chrome hit general availability on every paid plan, and memory now carries across chat and Cowork, on by default with sensitive topics excluded unless you opt them in. Local file access still requires the desktop app. September went further still: computer use now runs in the background — Claude clicks, types, and opens desktop apps while you work on something else (beta on Pro and Max, macOS and Windows apps; true background windows need macOS 15+). And Fable 5.1 fixed the thing users complained about most: Anthropic says its new safeguards produce 60% fewer false-positive blocks on security work, with biology safeguards firing 85% less often on benign requests — fewer wrongful refusals mid-task. Enterprise customers also got the first taste of Anthropic's new top tier: Claude Security, the code-scanning product in public beta, now runs on Claude Mythos 5 — the restricted Mythos-class model that sits above Opus in capability. Distribution is scaling the same way: the Claude Marketplace now lists a dozen partner products — Cursor, Gamma, Vercel, Factory, and CrowdStrike joined in early September — which enterprises can buy against an existing Anthropic spend commitment, with Anthropic handling the invoicing. And when things went wrong this summer — four incidents in which Claude models gained unauthorized access to real third-party systems during cybersecurity evaluations, traced to an eval environment misconfigured with live internet access — Anthropic's September 9 answer was to sign over 'wide-ranging access' to the nonprofit METR for an independent investigation, the first frontier lab to invite an outside audit of its own safety incidents. The incidents are the catch; the audit is the tell.

The extension story got a front door on September 23: Claude Marketplace, which Anthropic rebuilt as 'one place to discover plugins, agents, and services from our partners.' For a subscriber it means one directory of 2,000-plus connectors and plugins — Gmail, Drive, Slack, Notion, Microsoft 365, Figma, Atlassian — that you add on any plan, governed by a directory policy that bars third-party software from reading your chat history, memory, or files. The other two shelves aren't for you: buying Claude-powered products like Cursor or CrowdStrike's agent requires an enterprise spend commitment (that program has run in 'limited preview' since March), and the consulting roster is a referral list. What the Marketplace is and isn't has the full breakdown.

Claude pricing

The free tier is fine for evaluation but capped tightly enough that daily users will hit the wall. Pro at $20/month unlocks higher limits, Projects, Claude Code access, and — since September 22 — Opus 5.5, with Fable 5.1 available through usage credits. Anthropic raised the five-hour limits on Pro, Max, and Team with the launch and handed subscribers a rate-limit reset they can bank and spend when they choose. Max plans ($100 and $200/month) exist for heavy Claude Code users and researchers who want 5–20× Pro limits and every model including Fable.

Team and Enterprise plans add admin controls, SSO, and higher shared limits. Anthropic doesn't train on your conversations by default on any plan — a meaningful privacy distinction in this list. And if you run an academic or nonprofit research group: Anthropic opened 10,000 Team-plan seats for scientists in August — standard seats free, premium seats $15/month, for a year.