Claude Sonnet 5 Launches as Anthropic's Most Agentic Sonnet: What the Introductory Pricing Window Means for Routing Teams

Anthropic shipped Claude Sonnet 5 on June 30, 2026—and the model changes the routing calculus for AI engineering teams in three concrete ways: a new default model ID, a time-boxed introductory pricing window, and a redrawn cost-performance Pareto frontier between Sonnet and Opus.

What changed

Claude Sonnet 5 is available today across all Anthropic plans and on the Claude API via the model identifier claude-sonnet-5. It is now the default model for Free and Pro plans, replacing Sonnet 4.6.

Pricing for API callers:

Compared to Sonnet 4.6 ($3/$15), the introductory tier is a 33% discount on both input and output tokens—matching Sonnet 4.6's post-September price at the same capability level.

Key benchmark improvements over Sonnet 4.6:

• Higher scores on BrowseComp (agentic search) and OSWorld-Verified (computer use) at equivalent effort levels.
• Lower hallucination and sycophancy rates.
• Stronger resistance to prompt injection and malicious requests in agentic contexts.

Sonnet 5's performance approaches Opus 4.8 at higher effort settings, while costing 60–80% less per token.

Why it matters for AI engineering teams

Model ID migration is now active. Any system hard-coded to claude-sonnet-4-6 will stay on the old model. Teams routed to the generic claude-sonnet-latest or relying on a gateway's model alias system will automatically receive Sonnet 5. Verify which model ID your routing config resolves to.

The introductory window creates a two-month cost advantage. At $2/$10, Sonnet 5 is meaningfully cheaper than Sonnet 4.6 was at $3/$15—for the same or better output quality. Teams that have been holding off on upgrading from Sonnet 4.6 now have a cost incentive to migrate immediately rather than waiting for internal evaluation cycles.

Effort-level routing becomes the new tuning knob. Anthropic's BrowseComp charts show that Sonnet 5 at medium effort outperforms Sonnet 4.6 at maximum effort on several agentic tasks. This means the right routing policy for many teams is no longer "Sonnet for cost, Opus for quality" but rather "Sonnet 5 at calibrated effort for most tasks, Opus 4.8 only for tasks that demonstrably need it." That narrows the Opus use case.

Coding agent and multi-step automation workloads are the primary beneficiaries. Partner feedback cited in the announcement highlights: complex PR review completing without stalling, end-to-end automation of multi-system tasks, and sustained debugging without requiring re-prompts. If you run Cursor, Claude Code, or custom coding agents that previously required Opus for follow-through, Sonnet 5 is the first Sonnet worth retesting at full autonomy.

The router/operator angle

When to route to Sonnet 5 vs. Opus 4.8

The clearest decision framework based on the launch data:

• Sonnet 5 (high effort): Long-horizon coding tasks, computer use, multi-step tool chains, legal/research analysis, brownfield debugging. The model handles these at Opus-adjacent quality.
• Sonnet 5 (medium effort): Single-turn generation, summarization, structured output, classification. Best cost-performance.
• Opus 4.8: Tasks where Sonnet 5 at high effort still falls short in your evals—typically those requiring top-tier reasoning on novel problems or very long uninterrupted context chains. Retain as a fallback or premium tier, not a default.

Pricing window: when to lock in Sonnet 5

The August 31 deadline matters for teams with usage agreements or pre-negotiated rate structures. If you can shift volume to Sonnet 5 before September 1, the introductory $2/$10 rate applies. After that, Sonnet 5 reverts to $3/$15—identical to what Sonnet 4.6 cost at launch. There is no long-term discount advantage after August 31; the migration decision becomes purely capability-based.

Model naming and alias hygiene

Anthropic's model naming continues to add numeric suffixes (4.6 → 5). If your routing config uses exact version strings, pin to claude-sonnet-5 now and plan a review cycle before Anthropic ships the next increment. If you use claude-sonnet-latest, validate that your provider resolves this to Sonnet 5 and not a cached alias from a prior release.

What TheRouter users should watch or try

Teams routing through an AI gateway that supports Anthropic's API can update their provider model config to claude-sonnet-5 and begin capturing the introductory pricing window immediately. If you route with fallback chains—for example, Sonnet → Opus on error or timeout—consider whether Sonnet 5's improved agentic reliability reduces the frequency at which Opus fallback actually fires. Lower fallback rates translate directly to lower per-request cost.

Review your effort-level settings if your provider supports Anthropic's thinking parameter and reasoning_effort field—Sonnet 5's performance uplift is most pronounced at medium-to-high effort, so routing with a fixed low-effort setting underutilizes the model.

The Claude API models overview lists claude-sonnet-5 as the current model ID.