Claude Sonnet 5 is Anthropic's June 2026 workhorse, and inside Atlas it is the default choice for long agentic sessions. Sonnet 5 matches the 1M token window and 128K output ceiling of Opus 4.8 while costing $2 / $10 per Mtok (input / output), which is 60 percent below Opus 4.8 input pricing. The honest limit: on the hardest architectural reasoning, Opus 4.8 still pulls ahead and is worth the price delta.
Why is Claude Sonnet 5 the default model for long Atlas sessions?
Claude Sonnet 5, Anthropic's June 2026 workhorse, is the default for long agentic Atlas sessions because it matches the 1M token window and 128K output ceiling of Opus 4.8 while costing $2 / $10 per Mtok. Session length is exactly where per-token price compounds.
A long Atlas session is not one prompt. It is dozens of retrievals against a codebase indexed by AST declarations using tree-sitter, dozens of permission-gated tool calls, and a growing transcript that gets re-sent every turn. Input tokens dominate that bill, and Claude Sonnet 5 charges $2 per Mtok for them against Opus 4.8's $5. Because Sonnet 5 keeps the same 1M token context window, nothing about the session has to shrink to fit the cheaper price. You get the same working set for 60 percent less on input.
How much does Claude Sonnet 5 cost per million tokens?
Claude Sonnet 5 costs $2 / $10 per Mtok (input / output), making it the cheapest frontier-tier 1M context model in Anthropic's lineup. Against Claude Opus 4.8 at $5 / $25 per Mtok, that is 60 percent below Opus 4.8 input pricing for the same context window.
The pricing comparison only means something because the capability line holds. Claude Sonnet 5 keeps 128K output tokens, matching Opus 4.8, so it is not the bottleneck on large generated diffs. Atlas computes a unified diff for every file edit and surfaces it for approval before writing, and a model that can only emit half a patch forces you to stitch turns together. Sonnet 5 does not have that problem. What you actually give up at $2 / $10 per Mtok is peak architectural reasoning, not headroom.
Does Claude Sonnet 5 hold tool-call discipline across a long session?
Claude Sonnet 5 holds tool-call discipline across long sessions, which matters more in Atlas than in a chat UI. Every Atlas tool call is permission-gated against allow, ask, and deny rules, so a sloppy model burns turns on rejected calls and you pay for each one.
Tool-call discipline is an economic property in an agent, not a stylistic one. In Atlas the loop runs: retrieve, propose a tool call, clear the permission gate, act, observe. Every malformed call costs a full turn of input tokens plus your attention at the approval prompt. Claude Sonnet 5 stays well-formed deep into a session, which is what makes its $2 / $10 per Mtok rate real rather than nominal. A cheaper model that needs three attempts per call is not cheaper.
When should you upgrade from Claude Sonnet 5 to Opus 4.8?
Upgrade from Claude Sonnet 5 to Claude Opus 4.8 on the hardest architectural reasoning, where Opus 4.8 still pulls ahead and is worth the price delta. For everything below that bar, Sonnet 5 at $2 / $10 per Mtok produces the same unified diff for less money.
The upgrade is cheap to execute because Atlas lets you switch the active model and provider on the fly with favorites and recents. A practical pattern: run the session on Claude Sonnet 5, and when you hit a design question that genuinely needs frontier depth, open /models, switch to Anthropic's Opus 4.8 for that stretch, then come back. Atlas also drafts a plan in a read-only plan agent and asks before switching to a build agent, so the natural upgrade point is the planning pass, where a single expensive turn buys a better plan for the whole build.
How do you pin Claude Sonnet 5 as the everyday Atlas model?
Pin Claude Sonnet 5 as the everyday driver by setting "model": "anthropic/claude-sonnet-5" in atlas.json. Set ANTHROPIC_API_KEY in your environment first, or run atlas login and select Anthropic, then confirm the registry exposes it with atlas models anthropic.
Configuration and session control are separate in Atlas, and both work for Claude Sonnet 5. The atlas.json entry sets where every new session starts. The TUI dialog changes the current session: open /models, select Anthropic, then Claude Sonnet 5. Because Atlas keeps favorites and recents, a Sonnet 5 default plus a starred Opus 4.8 gives you a two-key toggle between the cheap workhorse and the frontier model without touching config again.
Setup
- 01Set `ANTHROPIC_API_KEY` in your environment, or run `atlas login` and select Anthropic.
- 02List what the registry exposes: `atlas models anthropic`.
- 03Open `/models` in the TUI and select Anthropic, then Claude Sonnet 5.
- 04Pin it as the everyday driver in atlas.json: `"model": "anthropic/claude-sonnet-5"`.
- 05Star Claude Opus 4.8 in favorites so you can switch up for the hardest architectural reasoning and back down again.
Frequently asked questions
- how much does claude sonnet 5 cost per million tokens
- Claude Sonnet 5 costs $2 / $10 per Mtok (input / output). That is 60 percent below Claude Opus 4.8 input pricing while keeping the same 1M token context window.
- what is claude sonnet 5's context window
- Claude Sonnet 5 has a 1M token context window and a 128K output ceiling, both matching Claude Opus 4.8.
- is claude sonnet 5 good enough for agentic coding
- Yes. Claude Sonnet 5 is Anthropic's June 2026 workhorse and holds tool-call discipline across long sessions, which is what long agentic Atlas sessions actually demand.
- claude sonnet 5 vs claude opus 4.8 for atlas
- Claude Sonnet 5 matches Opus 4.8's 1M token window and 128K output ceiling at $2 / $10 per Mtok instead of $5 / $25. Opus 4.8 still pulls ahead on the hardest architectural reasoning and is worth the price delta there.
- how do i set claude sonnet 5 as the default model in atlas
- Add `"model": "anthropic/claude-sonnet-5"` to atlas.json. For a single session, run `/models` in the TUI and select Anthropic, then Claude Sonnet 5.
- which api key does atlas need for claude sonnet 5
- Atlas needs `ANTHROPIC_API_KEY` set in your environment, or you can run `atlas login` and select Anthropic. Verify the model resolved with `atlas models anthropic`.
- can claude sonnet 5 emit a large multi-file diff in one turn
- Yes. Claude Sonnet 5 has 128K output tokens, matching Opus 4.8, so it is not the bottleneck on large generated diffs that Atlas surfaces for approval.
Try Atlas in your terminal
The terminal-native AI coding agent. Free core, single binary.
Install AtlasRelated guides
Run the Test Suite and Triage the Failures with Atlas in 2026
How to triage a failing test suite with Atlas in 2026: bash truncates at 2000 lines or 50 KB and saves the full log, then grep groups failures by root cause.
Atlas vs Poolside: Terminal AI Coding Agents in 2026
Compare Atlas and Poolside, two terminal AI coding agents in 2026. Atlas offers a free core and robust safety features, while Poolside provides open-weight Laguna models and on-prem deployment options.
Atlas for Zig: A Terminal-Native AI Coding Agent for build.zig Projects in 2026
Atlas is a terminal-native AI coding agent for Zig in 2026. It reads build.zig and comptime blocks, tracks your allocators, runs zig build test behind a prompt, and runs zig fmt.
Document a Module with a README Using Atlas (2026 Workflow)
How to document a module with a README using Atlas in 2026: the lsp tool's documentSymbol enumerates the real exports, read supplies the behavior, write emits the README.
Atlas for React in 2026
Adopt Atlas, the terminal-native AI coding agent, for React development in 2026. Enhance your workflow with intelligent code search, refactoring, and testing for React components and hooks.
Atlas vs Aider: Terminal AI Coding Agents in 2026
Comparing Atlas and Aider for 2026 developers. Atlas offers a rich TUI, permission-gated tool calls, and plugin support, while Aider provides a repo-map and commits every change as a discrete git revision.
Atlas for .NET: Your AI Coding Agent in 2026
Atlas empowers .NET developers in 2026 with a terminal-native AI coding agent. Securely integrate Atlas with ASP.NET Core for web APIs and services, leveraging local embeddings and robust safety features for efficient
Atlas for Deno: Terminal-Native AI Coding in 2026
Atlas is a terminal-native AI coding agent for Deno. Tighten --allow-all down to real permission scopes, add JSR deps, and run deno test and deno check in 2026.