Claude Sonnet 5 Just Became the Default in Claude Code — Near-Opus Agentic Coding at Less Than Half the Price

The Numbers Anthropic Just Put On The Table

On June 30, 2026, Anthropic shipped Claude Sonnet 5, the most agentic Sonnet model the company has released, and set it as the default model for every Free and Pro user on claude.ai from launch day — as well as the default in Claude Code for anyone running v2.1.197 or later. The number developers actually care about: Sonnet 5 scores 63.2% on SWE-bench Pro, up from Sonnet 4.6's 58.1%, closing most of the distance to Opus 4.8's 69.2% — while shipping at introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026 (rising to $3/$15 after), a fraction of what Opus 4.8 costs per token. For a model now running as the default across the highest-traffic Claude surfaces, that combination — near-flagship coding performance at mid-tier pricing — is the actual story.

A Full 1M-Token Context Window, No Long-Context Tax

The context window ships at 1 million tokens on the API with a 128K max output, and — notably — with no separate long-context premium the way some prior tiers charged extra once a conversation crossed a length threshold. For agentic workflows that read entire repositories, multi-file diffs, or long tool-call transcripts before producing a single edit, that is a meaningful operational change: the cost of a request no longer jumps just because your context passed 200K tokens. Combined with the default-high-effort configuration Sonnet 5 runs under on both the API and in Claude Code, this is a model tuned specifically for multi-step, tool-using, self-correcting agent loops rather than single-shot chat completions.

Where It Actually Wins — And Where Opus 4.8 Still Leads

Anthropic's own disclosed benchmark set draws a precise map of where Sonnet 5 sits. On Terminal-Bench 2.1, a coding-in-a-terminal evaluation, Sonnet 5 scores 80.4% against Sonnet 4.6's 67.0% and Opus 4.8's 82.7% — nearly closing that gap entirely. On OSWorld-Verified, a computer-use benchmark that tests whether a model can operate a GUI to complete a task, Sonnet 5 reaches 81.2%, up from 78.5%. On GDPval-AA v2, a knowledge-work benchmark, Sonnet 5 actually scores higher than Opus 4.8 — 1,618 versus 1,615. The pattern across all four disclosed benchmarks is consistent: Sonnet 5 does not beat Opus 4.8 on the hardest agentic coding tasks, but it gets close enough, on tasks broad enough, that the price difference becomes the deciding factor for most production workloads.

Why Discount Your Own Near-Flagship Model Right Now

The timing is not incidental. Anthropic filed confidentially with the SEC on June 1, 2026, for an IPO reportedly targeting a valuation near $1 trillion, and every product launch between now and that IPO functions, in part, as evidence for public-market investors: revenue growth, a credible path to profitability, and proof that enterprise customers genuinely prefer Anthropic's models over the competition. OpenAI, which raised $122 billion in March at an $852 billion valuation and has staggered its own GPT-5.6 rollout under US government pressure, is chasing the same enterprise budgets, alongside Google and Meta. Undercutting your own top-tier model's price with a near-flagship mid-tier release is a way to win volume and usage data ahead of an IPO roadshow, even if it compresses margin on paper — a strategy that only works if developers actually adopt the cheaper model in production, which is exactly what defaulting it into Claude Code and claude.ai is designed to accelerate.

What To Actually Do About It

If your team runs agentic workflows on Sonnet 4.6 today, the upgrade path is close to free: same API surface, same tool-use interface, better benchmarks, lower list price during the introductory window, and no action required if you are already pinned to the general "sonnet" alias rather than a dated model string. Claude Code users need v2.1.197 or later to get Sonnet 5 as default; teams on Team or Enterprise plans should confirm which model their org policy pins before assuming everyone was upgraded automatically. The more consequential decision is whether Sonnet 5's SWE-bench Pro and Terminal-Bench 2.1 scores are now close enough to Opus 4.8 that workloads currently routed to Opus for quality reasons should move down to Sonnet 5 for cost reasons — that is a per-workload evaluation, not a blanket switch, and it is worth re-running your own eval suite rather than trusting the published numbers alone.

Bottom Line

Claude Sonnet 5 is not a marginal point release. It is Anthropic's answer to the question of how much agentic capability you can get at mid-tier pricing — and the answer, measured across four independently disclosed benchmarks, is: most of it. Shipping as the default model in Claude Code and claude.ai from day one means the majority of Anthropic's developer traffic is now running through this model without anyone having to opt in. For developers and AI engineers deciding what to build agents on top of for the rest of 2026, the practical takeaway is simple: re-run your evals on Sonnet 5 before you keep paying Opus 4.8 prices for workloads it might now handle just as well.