Anthropic announced Claude Sonnet 5 on 2026-06-30, and Claude Code v2.1.197 (released 2026-06-30T17:56:37Z, the same day) made it the default model. The launch is the first Sonnet-line model that closes the agentic gap with Opus 4.8, ships a native 1M-token context window, and lands on introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, then $3/$15 per Mtok. Sonnet 5 is the new default for Free and Pro plans in Claude and the default model in Claude Code on update. The release ships with cyber safeguards on by default, a refreshed tokenizer, and rate-limit increases sized to absorb the higher token usage of the new effort-level controls.
The Sonnet 5 launch is the third major Anthropic event of the last three weeks. The June 12 Fable 5 distillation guardrails piece covered Anthropic's first public distillation disclosure naming DeepSeek, Moonshot AI, and MiniMax, and the June 13 Fable 5 / Mythos 5 US export control suspension piece covered the temporary worldwide suspension triggered by the June 12 US Commerce directive. The Fable 5 service resumes globally on July 1, the day after Sonnet 5 launches. The pairing is the same pattern Anthropic has used since the Project Glasswing report on April 7: one model line's restrictions land the same week as a smaller, cheaper model line's expansion.
What Sonnet 5 actually changes
The launch post positions Sonnet 5 as the model that lets the agentic AI era return to the Sonnet line. Claude Sonnet 3.5, 3.6, and 3.7 were the first models that showed strong coding and tool use; over the last year the clearest gains in agentic capabilities were on Opus-class models. Sonnet 5 narrows that gap, with performance close to Opus 4.8 at lower prices, on the two agentic evaluations Anthropic highlighted in the launch post: BrowseComp (agentic search) and OSWorld-Verified (computer use). The post shows cost-performance curves at different effort levels, with Sonnet 5 covering a much wider range of cost-performance options than Sonnet 4.6 and matching Opus 4.8 on some tasks at higher effort levels. Opus 4.8 is priced at $5/$25 per Mtok; Sonnet 5 at standard pricing ($3/$15) is 40% cheaper on both input and output, with a 60% cheaper effective price at the introductory tier.
The new tokenizer changes how Sonnet 5 processes text. The same input maps to 1.0 to 1.35x more tokens than the Sonnet 4.6 tokenizer, depending on content type. The change is similar to the one Anthropic introduced with Claude Opus 4.7. The introductory pricing is set so the transition to Sonnet 5 is roughly cost-neutral for users coming from Sonnet 4.6 at the same workload.
Claude Code 2.1.197 ships the model and a 1M-token context
The Claude Code 2.1.197 release notes are one line: "Introducing Claude Sonnet 5: now the default model in Claude Code, with a native 1M-token context window and promotional pricing of $2/$10 per Mtok through August 31. Update to version 2.1.197 for access." The 1M-token context window is the largest Anthropic has shipped on a Sonnet-line model, and it lands on the day Anthropic raises rate limits across Chat, Cowork, Claude Code, and the Claude Platform to absorb the higher token usage of the new effort-level controls. The new effort levels let Claude Code users select a cost-performance point per session, similar to the cost-performance curves in the launch post.
The 2.1.197 release follows the v2.1.196 release the day before (2026-06-29T23:27:32Z), which added org-default models, clickable Cmd/Ctrl-click file attachments that open in Finder/Explorer, and a security fix for .mcp.json servers. The MCP fix is worth flagging on its own: claude mcp list and claude mcp get no longer spawn servers that a repo self-approved via a committed .claude/settings.json. Untrusted workspaces now show ⏸ Pending approval and require an explicit user gesture to start the server, which closes a trust gap that the prior behavior quietly relied on the developer to notice. The same release also fixed mid-turn crash recovery for Remote sessions (interrupted sessions now auto-resume on the next worker), rate-limit telemetry over-counting under parallel requests, and /context showing 0 tokens for all tool groups on Bedrock.
Safety, cyber safeguards, and the system card
Anthropic's pre-deployment safety evaluations found Sonnet 5 to be an overall improvement on Sonnet 4.6. The three concrete gains: lower hallucination and sycophancy rates, lower misaligned-behavior rate on the automated behavioral audit, and tighter agentic safety, with the model better at refusing malicious requests and resisting hijack attempts in prompt-injection attacks. The one regression: on the same audit, Sonnet 5 scores somewhat higher on misaligned behavior than Opus 4.8 and Claude Mythos Preview. The launch post and the system card both state the trade-off explicitly.
On cyber, Sonnet 5 was never deliberately trained on cybersecurity tasks. It can perform some routine, non-harmful cyber work, but on the Firefox 147 exploit-development evaluation developed in collaboration with Mozilla, neither Sonnet model could develop a working exploit (both at 0.0%); Sonnet 5 has a slightly higher partial-success rate than Sonnet 4.6, which Anthropic attributes to general-intelligence improvements rather than specific training. Both Sonnet models have substantially poorer cyber capabilities than Opus 4.8 and Mythos 5. Because Sonnet 5 is somewhat stronger than Sonnet 4.6 on these tasks, it ships with cyber safeguards on by default, the same set as Opus 4.7 and 4.8. The Fable 5 safeguards, which block a wider range of cybersecurity tasks, are stricter.
The full Sonnet 5 system card is published alongside the launch post and reports a broader set of evaluations in detail. The Cyber Verification Program is available on the native Claude Platform, Claude Platform on AWS, and Claude in Microsoft Foundry, and is coming soon on Claude in Google Vertex.
Why this matters for TypeScript and web developers
The Claude Code audience is the one that gets the immediate effect. Every new Claude Code session on v2.1.197+ runs on Sonnet 5 by default, with a 1M-token context window that covers an entire medium-sized monorepo for an agentic edit, and a per-token cost that is unchanged from Sonnet 4.6 in practice (the introductory tier is set to absorb the 1.0 to 1.35x tokenizer inflation). The cost-performance point of the new effort-level controls means a developer can pick a cost-effective model for routine refactors and a more capable model for the long-horizon tasks, in the same session, without switching out of Claude Code. The MCP trust fix in v2.1.196 closes a quiet supply-chain risk for teams that have been running Claude Code against repos with a committed .claude/settings.json, and the 2.1.197 release lands a model that is, on Anthropic's own numbers, the first Sonnet-line model that can stand in for Opus 4.8 on the agentic work the Claude Code audience does every day.
