Claude Code release notes: 2.1.196-2.1.197 make Sonnet 5 the default

Claude Code 2.1.197 makes Sonnet 5 the default model in Claude Code. That sounds like a routine changelog line, but for teams already using Claude as a coding agent it is more concrete: larger context, lower Sonnet pricing during the launch period, and a sharper question about who gets to decide model choice inside the organization.

Source: Claude Code changelog

Source: GitHub release v2.1.197

Source: Anthropic: Introducing Claude Sonnet 5

Claude Code release notes: Sonnet 5 becomes the default in 2.1.197

A coding agent is an AI assistant that can read a codebase, suggest changes, and work across files and tools instead of only answering in chat. In Claude Code 2.1.197, Claude Sonnet 5 is now the default model. Anthropic describes Sonnet 5 as its most agentic Sonnet model so far, with better planning, tool use, coding, knowledge work, computer use, and agentic search than Sonnet 4.6.

The practical value is the combination. Claude Code gets a default model with a native 1-million-token context window. That does not make every task better by itself, but it makes larger codebases, longer requirement documents, and more logs easier to keep in the same workspace. Anthropic also lists introductory API pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026.

Source: Anthropic: Introducing Claude Sonnet 5

Source: npm registry for @anthropic-ai/claude-code

2.1.196: model governance and more reliable background agents

Release 2.1.196 is less flashy, but important for real operations. It adds support for organization default models. Admins can set the default model in the organization console, and Claude Code shows "Org default" or "Role default" in /model when the user has not chosen one.

The same release makes agent work easier to review. claude mcp list and claude mcp get should no longer start .mcp.json servers that a repository self-approved through a committed .claude/settings.json. In untrusted workspaces, Claude Code shows ⏸ Pending approval instead. MCP, the Model Context Protocol, is a way to connect an AI assistant to external tools and data sources with defined boundaries.

Background agents also get several reliability fixes: conversations should not be deleted when a transcript probe misreads a file, stopped or restarted processes should be handled better, and long-running background commands should survive stops, restarts, and updates more reliably, including on Windows.

Source: GitHub release v2.1.196

Source: Claude Code changelog

What this means for Nordic teams

This is not only a model upgrade. It is a good moment to decide how Claude should become part of the work environment.

For a smaller team, that can mean three simple decisions:

• Which model should be the default for normal code and document work?
• When should someone choose a different model for cost, quality, or speed?
• Which MCP servers, file paths, and environment variables may an agent use without asking again?

The last point is where the integration becomes real. Use scoped permissions, secrets in env vars or secret managers, redact sensitive output before it is logged, add approval gates for production systems, and keep audit logs for agent runs. Then Claude can do useful work without turning everything into a manual chat exercise.

Try this prompt this week

Human step: use an environment where Claude Code 2.1.197 is available, or collect your current Claude Code settings, model policy, MCP configuration, and background-agent flows. Do version checks and any updates outside the prompt.

Read our Claude Code model policy, MCP configuration and latest agent logs.
Compare them with 2.1.196-2.1.197: Sonnet 5 as default, org default models, MCP approval and background agents.
Suggest the three smallest changes we should test this week.
Mark what needs human approval before an agent may read files, use tools or write logs.
Answer as a short operations receipt with open questions last.

Good output should include:

• one clear recommendation for the default model and when to override it
• one concrete check of MCP servers or repository settings
• one note about which agent logs should be kept and redacted
• one small test task where Sonnet 5's larger context actually matters

Hammer angle

This is best treated as a Tool Forge question: make Claude useful in real workflows, but build model choice, permissions, secrets, approvals, and logs into the solution from the start. Then the team gets more than a stronger chat model. It gets a workspace it can follow up on.