Anthropic Claude Sonnet 5: Most Agentic, Near-Opus Quality

Anthropic Launches Claude Sonnet 5 on 30 June 2026

On 30 June 2026, Anthropic launched Claude Sonnet 5, the company's most capable mid-tier model to date and its most aggressively positioned product for agentic workloads. The model went live immediately as the default on claude.ai for every Free and Pro subscriber, replacing Claude Sonnet 4.6 without a staged rollout. Simultaneously, it became available through the Anthropic API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry. Unlike Claude Fable 5, which launched under export-control restrictions in June 2026, Sonnet 5 carries no access gates and is available globally across all subscription tiers — Free, Pro, Max, Team, Enterprise, and Claude Code — from day one.

What Changed From Sonnet 4.6

Claude Sonnet 5 is built around autonomous, multi-step task completion at a quality level that previously required Anthropic's larger Opus-class models. Anthropic describes it as capable of finishing complex tasks that Sonnet 4.6 could not complete, and the model self-verifies its outputs without being explicitly prompted — a shift that removes one layer of developer overhead in agentic pipelines where external validation loops would otherwise be required.

The benchmark scores Anthropic published at launch confirm the improvement. On SWE-bench Pro, the standard benchmark for autonomous software engineering, Sonnet 5 scores 63.2 per cent, up from 58.1 per cent for Sonnet 4.6. Claude Opus 4.8 still leads the Anthropic family at 69.2 per cent, but the gap between the mid-tier and the flagship has narrowed considerably. On OSWorld-Verified, a computer-use benchmark, Sonnet 5 posts 81.2 per cent against Sonnet 4.6's 78.5 per cent. The Terminal-Bench 2.1 result is the most striking: Sonnet 5 reaches 80.4 per cent versus Sonnet 4.6's 67.0 per cent, a 13.4-percentage-point improvement on long-horizon shell and system interactions.

Anthropic has also improved safety characteristics relative to the previous model. Sonnet 5 demonstrates better refusal of malicious requests, greater resistance to prompt injection attacks, and lower rates of hallucination and sycophancy. For teams deploying agents against real systems or processing sensitive data, these properties meaningfully reduce the risk of a misaligned or manipulated agent taking unintended actions at decision points in an automated workflow.

Pricing and Context Window

Anthropic set an introductory price of 2 dollars per million input tokens and 10 dollars per million output tokens, valid through 31 August 2026. After that date, pricing moves to 3 dollars per million input tokens and 15 dollars per million output tokens. The two-month introductory window gives teams a practical opportunity to build, evaluate, and scale on Sonnet 5 before committing to the standard rate.

Sonnet 5 ships with a one-million-token context window and 128,000 maximum output tokens. Adaptive thinking — Anthropic's optional extended reasoning mode — is supported, allowing the model to spend additional compute working through hard problems before generating a final response when the task warrants it.

Model ID and Availability Across Platforms

The model identifier is claude-sonnet-5. It is immediately available with no waiting period on all Claude plans, through Claude Code for terminal-based workflows, and through Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry for teams deploying via cloud marketplaces. No geographic restriction applies to Sonnet 5 at launch — a meaningful contrast to the US export-control directive that has kept Claude Fable 5 suspended for non-US users since 12 June 2026.

What Sonnet 5 Means for Indian Development Teams

For Indian engineering teams, the timing of Sonnet 5's arrival is practically significant. Claude Fable 5, Anthropic's most capable general-purpose model, remains suspended under US export controls as of early July 2026. Sonnet 5 fills that gap as the strongest Anthropic model currently available without access restrictions, and it does so at a price point — 2 dollars per million input tokens during the introductory period — that keeps agentic workloads economically viable for Indian teams where API spend scales directly with customer-facing deployment volume.

The Terminal-Bench 2.1 gains are particularly relevant for Indian software services firms running agentic pipelines against cloud infrastructure, CI/CD systems, or enterprise databases. Teams using Claude for multi-file coding agents, automated documentation, enterprise support workflows, or long-horizon research tasks will find Sonnet 5's self-verification behaviour reduces the human oversight required at each agent step, improving throughput without proportionally increasing error rates. The one-million-token context window also supports common enterprise use cases in India's legal, financial, and compliance software verticals, where document lengths regularly exceed what smaller-context models can handle in a single call.

The Bottom Line

Anthropic launched Claude Sonnet 5 on 30 June 2026 as an immediate global release across all subscription tiers with model ID claude-sonnet-5 and no access restrictions. Benchmark scores of 63.2 per cent on SWE-bench Pro, 81.2 per cent on OSWorld-Verified, and 80.4 per cent on Terminal-Bench 2.1 place it near Opus 4.8 quality for agentic and computer-use tasks. Introductory pricing of 2 dollars per million input tokens and 10 dollars per million output tokens runs through 31 August 2026, then rises to 3 and 15 dollars respectively. For Indian teams navigating restricted access to Fable 5, Sonnet 5 is the most capable Anthropic model currently available without conditions, paired with a one-million-token context window and adaptive thinking support.