Claude Sonnet 5.5 joins GitHub Copilot

GitHub made Anthropic’s Claude Sonnet 5.5 generally available in GitHub Copilot on September 28, 2026, according to its GitHub release note. GitHub’s evaluations found that the model matched Sonnet 5 on coding tasks while completing them with fewer steps, tokens, and tool calls.

Anthropic released Sonnet 5.5 days after the higher-end Opus 5.5, extending the Claude 5.5 family from complex, long-horizon work to more tightly scoped engineering tasks. Sonnet’s lower price and reduced tool use make it the volume-oriented option for coding agents that may run dozens of turns inside a repository.

Fewer calls cut agent overhead

GitHub reports that Sonnet 5.5 reached Sonnet 5-level coding results with fewer model steps, tokens, and tool calls. Each avoided turn can remove model latency, shell execution time, API calls, and token charges from an agent run.

Anthropic says the model generates output more than 30% faster and can cost up to 30% less per task in its testing. The per-token rates remain unchanged from Sonnet 5, so the claimed savings come from lower token consumption and fewer interactions. Actual latency and cost will vary with repository size, tools, prompts, and provider limits.

Lovable co-founder and CTO Fabian Hedin reported similar findings from the company’s evaluations. Sonnet 5.5 used about one-third fewer tool calls and roughly half as many shell executions to finish coding jobs, reductions that can shorten multi-step refactors and debugging sessions.

Sonnet tops Opus on one terminal test

Anthropic reports a 70.6% score for Sonnet 5.5 on Terminal-Bench 4.0, compared with 10.3% for Sonnet 5 and 66.4% for the higher-priced Opus 5.5. The result places the mid-tier model ahead of Anthropic’s flagship within that specific benchmark harness.

Terminal-Bench evaluates whether an agent can complete shell-driven tasks from start to finish, including planning, command execution, file changes, and recovery from errors. It therefore tests the full agent loop rather than isolated code generation.

Benchmark scores depend on the prompts, scaffolding, tools, time limits, and task set used in the evaluation. Teams considering a migration can compare the models on representative repositories and record completion rate, wall-clock time, token use, tool calls, and review effort.

Migration details that affect requests

• Adaptive thinking is enabled by default. Sonnet 5.5 decides how much internal reasoning to use during a task. Applications that explicitly disabled thinking for Sonnet 5 need to migrate to the new between_tools setting before changing model IDs. Opus 5.5 already rejects requests that disable thinking entirely.
• The underlying limits remain unchanged. Sonnet 5.5 supports a 1 million-token context window, a 128,000-token output ceiling, and a June 2026 knowledge cutoff. Individual products and cloud providers may enforce lower limits.
• Additional safeguards apply. This is the first Sonnet release with Anthropic’s frontier-style cybersecurity controls and classifiers designed to block attempts to extract protected internal reasoning. Anthropic says its cybersecurity capabilities are comparable to those of Opus 5.
• Cloud access extends beyond Copilot. Developers can use the model through Anthropic, Amazon Web Services, Google Cloud, and Microsoft Azure. Anthropic also offers zero-data-retention support for eligible API usage.

Enable it across Copilot

1. Open the model picker in a supported IDE, the Copilot CLI, the coding agent, GitHub Mobile, or github.com.
2. Select Claude Sonnet 5.5 for the session or task.
3. For Business and Enterprise accounts, confirm that an administrator has allowed the model under the Copilot model policy.

GitHub is propagating the release gradually, so eligible accounts may receive the model at different times. New models are enabled by default for Business and Enterprise accounts unless an administrator disables them.

Usage-based Copilot billing applies Anthropic’s provider list rates with a 1x Copilot multiplier. At those rates, input costs $2 per million tokens and output costs $10 per million tokens.

Sonnet targets bounded work

Anthropic positions Sonnet 5.5 for feature implementation, bug fixes, focused refactors, code review, and polished documents, slides, and spreadsheets. These workloads have a defined goal and enough complexity to benefit from planning and tool use.

Opus 5.5 remains the option for open-ended research, long-horizon reasoning, and complex tasks whose scope may change during execution. Sonnet 5.5 fits high-volume engineering workflows where completion time, tool-call count, and token use directly affect operating cost.