What are the verified coding and reasoning benchmarks for Claude Sonnet 4.5?

Claude Sonnet 4.5 achieves SWE-bench Verified: 72.8% (Extended Thinking), LiveCodeBench: 64.1%, AIME: 84.5%. Verified sources: Claude Sonnet 4.5 Architecture and Capabilities (Anthropic).

Does Swarm isolate Claude Sonnet 4.5 code edits from other agents?

Yes. Swarm provisions isolated Git worktrees for every session and subagent, preventing file race conditions and keeping your main working tree clean until verification passes.

Can Claude Sonnet 4.5 be assigned to specific subagent roles?

Yes. In ~/.swarm/config.json, you can configure Claude Sonnet 4.5 as Primary/Planner (DAG task decomposition), Coder (implementation in isolated worktrees), or Finder (read-only symbol search).

Claude Sonnet 4.5 — Specs, Context & Capabilities

Q: What is the recommended Swarm multi-agent topology for Claude Sonnet 4.5?

Swarm recommends Claude Sonnet 4.5 as Primary / Planner (Tier 1: Next-Gen Systems Architect). Supreme capability in multi-file dependency analysis. Perfect as the Lead Planner for complex full-stack refactors and large database migrations. Best paired with: google/gemini-3.8-flash for rapid subagent execution.

Q: How do I run Claude Sonnet 4.5 with Swarm?

Configure Claude Sonnet 4.5 via CLI using: swarm ctl model set --provider anthropic --model anthropic/claude-sonnet-4-5-20250929, or switch models interactively inside the terminal TUI with /models.

Claude Sonnet 4.5 Benchmarks & Agentic Accuracy

Verified primary research & developer preference metrics

Primary Sources Verified

SWE-bench Verified

72.8%

Extended ThinkingAnthropic Technical Preview ↗

LiveCodeBench

64.1%

ThinkingLiveCodeBench v2 ↗

LMSYS Coding Elo

1410Elo

Coding ArenaLMSYS Arena ↗

AIME 2024 / 2025

84.5%

ThinkingAnthropic Technical Preview ↗

Pricing ($ / 1M tokens)

$3/ $15

$0.35 cached readStandard API Pricing

What This Means in Real Software Development

Pushes SWE-bench Verified to 72.8%, resolving nearly 3 out of 4 real GitHub issues with zero human feedback.
Deep multimodal reasoning enables interpreting complex visual UI regressions, architectural diagrams, and performance flamegraphs directly.
Maintains predictable $3.00/M input pricing while expanding internal thinking depth and compiler-level AST precision.
Optimized for deterministic DAG workflow execution inside Swarm Task Programs.

Swarm Multi-Agent Topology

Tier 1: Next-Gen Systems ArchitectRole: Primary / Planner

Best Paired With: google/gemini-3.8-flash for rapid subagent execution

Supreme capability in multi-file dependency analysis. Perfect as the Lead Planner for complex full-stack refactors and large database migrations.

Git Worktree Isolation Advantage:

Integrates with Swarm worktree isolation to test full build pipelines on disposable branch checkouts.

What Developers & Practitioners Are Saying

Real-world field observations, production strengths, and developer consensus

Field Consensus

"Highly capable intermediate Sonnet revision, offering improved tool reliability and tighter error-recovery loops."

Verified Real-World Strengths

72.8% on SWE-bench Verified
Refined multi-file patch precision with low hallucination
Solid balance of cost and execution speed

Production Caveats & Gaps

Largely superseded by Claude 3.7 Sonnet for frontier reasoning tasks

Primary Sources & Methodology Citations

Claude Sonnet 4.5 Architecture and Capabilities — Anthropic (September 2025)

Claude Sonnet 4.5 model overview

Claude Sonnet 4.5 is listed in the public Swarm catalog through Anthropic. This page summarizes the verified catalog fields used by Swarm without claiming real-time access for every provider account.

Provider: Anthropic
Family: Claude
Context window: 1M tokens
Thinking support: off
Availability status: Available
Catalog ID: anthropic/claude-sonnet-4-5-20250929

File input
Image input
Text input
Text output
Reasoning
Structured output
JSON
Batch

Run Claude Sonnet 4.5 in Swarm

Verified CLI workflow & multi-agent architecture

Quickstart Guide →

bash — swarm run

$swarm ctl model set --provider anthropic --model anthropic/claude-sonnet-4-5-20250929

# Or switch interactively inside the terminal TUI:

>/models

Subagent Role Matrix

Assign Claude Sonnet 4.5 to specialized system roles in ~/.swarm/config.json:

Primary / Planner: Orchestrates full task DAGs and multi-stage pipelines.
Coder Subagent: Generates code and unit tests inside private Git worktrees.
Finder Subagent: Performs read-only code search, symbol discovery, and mapping.

System Agent Roles →Why Swarm Architecture →