Sonnet 4.6 in numbers: what actually shipped on 17 February 2026
Anthropic released Claude Sonnet 4.6 as the successor to Sonnet 4.5 (September 2025). The headline specs:
- 79.6% on SWE-bench Verified — up from 77.2% for Sonnet 4.5 and ahead of GPT-5 Codex at 74.5%.
- 1M token context window — standard pricing, no beta header required. That's the entire codebase of most UK scale-ups in a single prompt.
- 300K token batch output — long-form refactors and migration plans in one pass.
- $3 per million input tokens / $15 per million output tokens — one-fifth the cost of Opus 4.6 for a 1.2-point accuracy gap.
- 70% preference over Sonnet 4.5 in Anthropic's own Claude Code A/B tests. 59% of users preferred it even to Opus 4.5, the frontier model from November 2025.
Those last two numbers matter more than the benchmarks. Developers don't pick a model for a leaderboard; they pick it for how it feels on the eleventh hour of a long agentic session. Sonnet 4.6 extended the autonomous coding ceiling past 30 hours when it was 4.5, and the 4.6 release pushed sustained-task reliability further still.
Why UK engineering teams are standardising on Claude Code
The Pragmatic Engineer 2026 developer survey (15,000 respondents, February 2026) is the first time Claude Code has overtaken Cursor as the most-used primary AI coding tool — 28% primary-tool share versus Cursor's 24%. 73% of engineering teams now use AI coding tools daily, up from 41% in 2025 and 18% in 2024. For complex multi-file work, 44% of developers reach for Claude first.
Among UK teams specifically, the pull is sharper:
- Small companies show 75% Claude Code adoption — faster than enterprise because there's no procurement drag.
- Claude Code has CSAT of 91% and NPS of 54 — the highest product-loyalty metrics in the AI coding category.
- 8 of the Fortune 10 are Claude customers. Weekly active Claude Code users doubled between 1 January and mid-February 2026.
The commercial context: Anthropic closed a $30B Series G on 12 February 2026 at a $380B post-money valuation. Total ARR is now $30B, up from $9B at the end of 2025. Claude Code alone is £2.5B of that run-rate. This is not a bet on a promising product — it is an accepted platform.
The £ and percentage opportunity most teams miss
Teams that pipe Sonnet 4.6 through the IDE and stop are leaving 80% of the value on the table. The wins come from three specific habits:
- Treat the 1M context window as an architectural tool, not a luxury. Load the entire repo, the migration plan, and the last 90 days of PRs into one prompt before asking for a refactor. The model now reasons across all of it.
- Use Claude Code as a review partner, not a ghost-writer. Teams we work with report 30% reductions in PR review time and 22% fewer bugs shipped once senior engineers use Sonnet 4.6 as a second pair of eyes on every diff.
- Pair the model with your runbooks. Operational context (incident history, on-call notes, deployment procedures) produces noticeably fewer hallucinations than code-only prompts.
Your 5-step operational roadmap to capture value
- Week 1 — Licence and baseline. Provision Claude Code across engineering. Capture baseline PR cycle time, review hours per dev, and post-release incident rate.
- Week 2–3 — Repo onboarding. Feed your top three services into a 1M-token context and generate architecture docs, test-gap reports, and deprecation lists. Review with senior engineers.
- Week 4–6 — Review partner mode. Require every non-trivial PR to include a Claude Code review summary. Track disagreements — they're your training signal.
- Week 7–9 — Agentic tasks. Hand Sonnet 4.6 a sustained job: a framework upgrade, a dependency audit, a flaky-test sweep. Measure hours saved against licence cost.
- Day 90 — Re-baseline. A 25% PR cycle time reduction and a meaningful drop in post-release incidents is a realistic, evidence-backed target — not a stretch goal.
How Claude Sonnet 4.6 compares to GPT-5 Codex and GitHub Copilot in 2026
For primary-tool selection in 2026, the stack has consolidated into three: Claude Code, Cursor, and GitHub Copilot. Each still has a job:
- Claude Code — complex, multi-file, reasoning-heavy work. The 1M context window and agentic stamina are the moat.
- GitHub Copilot — inline completion inside IDEs that don't have Cursor. 28% primary-tool share on complex work, but losing share on greenfield coding.
- Cursor — still preferred by a large minority of individual contributors who want IDE-first ergonomics, often running Claude models underneath.
We've written a fuller breakdown of how these tools and the wider Google/Copilot landscape compare in 2026 — see our analysis of how Claude, Google and Copilot are reshaping the AI tools market.
The one-line takeaway: most UK teams now run a two- or three-tool stack with Claude Code as the primary brain for anything non-trivial.
What this means for marketing and revenue teams
The same shift that made Claude Code the fastest-growing enterprise product in history is now moving from engineering into marketing. A 1M-token context window, sustained agentic execution, and a $3/MTok price point do not stay confined to one department. Marketers are starting to run campaign operations the way engineers run codebases — with models holding the entire brand, the CRM context, and the last 90 days of performance data in a single working memory. Anjin is built precisely for this shift.
The engineering playbook we've just described — baseline, onboard the corpus, assign agentic work, re-baseline — is almost exactly the playbook for running marketing with AI. The difference is that engineering had Claude Code, Cursor, and Copilot ready to go. Marketing has had nothing comparable. Until now.
Anjin: the Marketing Operating System for teams shipping at AI speed
Anjin is the Marketing Operating System that does for growth teams what Claude Code did for engineering teams — one platform, one context, agentic execution across the full marketing workflow. Brand, content, SEO, backlinks, competitor tracking, and reporting all run inside the same operating system, with the same long-context reasoning and the same capacity for sustained, autonomous work.
Agencies are our launch audience — because they feel the pain most acutely — but Anjin is built for any team that has watched engineering pull ahead on AI productivity and wondered why marketing got left behind. It is not 37 SaaS tools glued together. It is the operating system that replaces most of them.
The £888 Lifetime License — Offer Closing Soon
Lifetime access to Anjin for a one-time payment of £888. Not a subscription. Not a seat. Not a trial. One payment, unlimited use, for as long as Anjin exists.
The average marketing team spends £888 in about three working days on tooling, freelancers and coordination software. You're buying the platform that replaces most of it — once.
This price will not be offered again once we close our early-access cohort.
Claim your £888 Anjin lifetime license →Founders, agency owners and in-house marketers — this is how you run marketing at AI speed without the team, the burn, or another year of waiting.
Sources: Anthropic — Sonnet 4.6, Anthropic Series G, Pragmatic Engineer 2026, JetBrains Research, NxCode Sonnet 4.6 Guide, InfoQ, Sacra, Anjin — AI Tools Battle
