If you use Claude Code on a Pro, Team Standard, or Enterprise plan, you are already running Claude Sonnet 5. It became the default model on June 30, 2026, and most users never noticed — which is exactly the point.
Sonnet 5 is the model that powers the majority of Claude Code sessions worldwide. Understanding what it is, how it compares to the Opus and Fable tiers, and how its pricing works will save you money and help you pick the right model for each task.
What Claude Sonnet 5 Is
Claude Sonnet 5 is Anthropic’s latest mid-tier model, positioned below Opus 5 and Fable 5 in capability but above Haiku in intelligence. The headline: it delivers performance close to Opus 4.8 at roughly half the cost.
Key facts:
- Native 1M token context window — the same massive context as Opus 5, enough for entire codebases
- Adaptive thinking on by default — the model reasons through problems step by step without you asking
- $2 per million input tokens, $10 per million output tokens through August 31, 2026 (introductory pricing)
- Standard pricing after August 31: $3/$15 per million tokens
- Available on all plans: Free, Pro, Max, Team, Enterprise, and through the API
That promotional pricing matters. At $2/$10 per MTok, Sonnet 5 is cheaper per token than Sonnet 4.6 was at its standard rate, despite being a significant capability upgrade. The token counts may shift slightly due to a new tokenizer (roughly 1.0–1.35× more tokens per input), but Anthropic set the introductory price to make the transition roughly cost-neutral.
Why It Matters for Coding Agent Users
For most developers using Claude Code, Sonnet 5 is not an optional upgrade — it is the model you already have. Here is why that is good news.
1. Agentic performance that used to require Opus
Sonnet 5 narrows the gap with Opus-class models on the capabilities that matter for coding agents: reasoning, tool use, sustained multi-step work, and self-correction. Anthropic’s own evaluations show Sonnet 5 matching or approaching Opus 4.8 on agentic tasks at medium and high effort levels.
Early-access partners reported concrete improvements:
- “We handed Claude Sonnet 5 a two-part job — update Salesforce account tiers, send a launch announcement — and it finished end to end. That used to stall halfway.”
- “I asked Claude Sonnet 5 to investigate a bug. Unprompted, it wrote a reproducing test, implemented the fix, then stashed it to confirm the bug came back without the change. All in a single pass.”
- “Claude Sonnet 5 handled the full range of coding tasks we tested it on, while resolving more issues.”
2. The 1M context window changes what you can do
With a native 1M token context, Sonnet 5 can hold an entire large codebase in working memory. This means fewer context compactions, less lost state, and better performance on tasks that span many files. You do not need to upgrade to Opus for the context window anymore.
3. Adaptive thinking without extra configuration
Sonnet 5 enables adaptive thinking by default. The model reasons through complex problems internally before responding, which improves code quality and reduces back-and-forth. No /effort command needed — it just works out of the box.
4. Safer than its predecessor
Anthropic’s safety evaluations found Sonnet 5 has lower rates of hallucination, sycophancy, and undesirable behavior compared to Sonnet 4.6. It is also better at refusing malicious requests and resisting prompt injection attacks. For coding agents running autonomously, this means fewer surprises.
How Sonnet 5 Compares to Other Claude Models
Here is where Sonnet 5 fits in the current Claude model lineup:
| Model | Price (input/output per MTok) | Context | Best For |
|---|---|---|---|
| Haiku | ~$0.25/$1.25 | 200K | Simple tasks, quick edits |
| Sonnet 5 | $3/$15 (promo: $2/$10) | 1M | Daily coding, most tasks |
| Opus 5 | $5/$25 | 1M | Complex reasoning, hardest tasks |
| Fable 5 | Custom | 1M | Frontier research, specialized |
The practical decision tree:
- Everyday coding, debugging, refactoring, writing tests: Sonnet 5. This is what the default model is for.
- Complex architectural decisions, multi-file refactors, novel problem solving: Consider Opus 5, which has stronger reasoning at a higher price.
- Tasks requiring the absolute maximum capability regardless of cost: Fable 5, with its 30-day data retention caveat.
For the vast majority of Claude Code sessions, Sonnet 5 is the right choice. Opus 5 adds value on the hardest tasks, but the cost-per-task ratio favors Sonnet 5 for routine work.
How to Use Sonnet 5 in Claude Code
If you are on a Pro, Team Standard, or Enterprise plan, Sonnet 5 is already your default. You do not need to do anything.
To verify or switch manually:
# Check which model is active
/model
# Switch to Sonnet 5 explicitly
/model claude-sonnet-5
# Switch back to the account default
/model defaultIf you are on Max, Team Premium, or Enterprise pay-as-you-go, your default is Opus 5. To switch to Sonnet 5 for lighter tasks:
/model claude-sonnet-5This saves your selection, so future sessions start with Sonnet 5 until you change it again.
For API Users
The model identifier is claude-sonnet-5:
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "anthropic-version: 2023-06-01" \
-d '{
"model": "claude-sonnet-5",
"max_tokens": 4096,
"messages": [{"role": "user", "content": "Explain this error"}]
}'Effort Levels
You can tune Sonnet 5’s reasoning depth:
# Default — balanced speed and quality
/effort medium
# More thorough reasoning for complex tasks
/effort high
# Maximum reasoning (closer to Opus-level quality)
/effort xhigh
# Fast responses for simple tasks
/effort lowHigher effort = more tokens consumed = better results on hard problems. For most daily coding work, the default medium effort is fine.
Pricing: The August 31 Deadline
The most important thing to know about Sonnet 5 pricing right now: the introductory rate ends August 31, 2026.
| Period | Input (per MTok) | Output (per MTok) |
|---|---|---|
| Now through August 31 | $2 | $10 |
| After August 31 | $3 | $15 |
If you are doing high-volume API work, the 30 days remaining on the promotional rate represent a meaningful cost saving. A million input tokens at $2 vs $3 is a 33% difference. For teams processing thousands of requests per day, this adds up fast.
Plan your heaviest API usage before August 31 if possible.
What Opus 5 Users Should Know
If you are running Opus 5 (the default for Max and Team Premium), Sonnet 5 is worth considering for lighter tasks. Here is the practical split:
Use Opus 5 for:
- Complex multi-file refactors
- Architectural decision-making
- Tasks where you need the strongest reasoning available
- Security-sensitive work (Opus has lower classifier intervention than Fable)
Switch to Sonnet 5 for:
- Routine bug fixes
- Writing tests
- Code reviews on focused changes
- Documentation updates
- Any task where speed and cost matter more than maximum reasoning depth
You can switch mid-session with /model claude-sonnet-5 and switch back with /model opus when you hit a harder problem.
How Other Agents Handle Sonnet 5
The ecosystem adopted Sonnet 5 quickly:
- Pi added Sonnet 5 support starting with v0.81.0 (July 21), including adaptive thinking and prompt caching. If you run Pi alongside Claude Code, you can use Sonnet 5 in both.
- Cursor supports Sonnet 5 through its model picker. Cursor’s “Auto mode” powered by Cursor Router can select Sonnet 5 for appropriate tasks.
- OpenAI Codex does not use Claude models natively, but the broader trend of “smart default models at mid-tier pricing” is reshaping how all agents think about model selection.
Common Questions
Will Sonnet 5 replace Opus 4.8 entirely?
Yes, for most users. Sonnet 5 is the new default, and Opus 4.8 is being phased out of fast mode. If you explicitly select Opus 4.8, you will still get it, but there is little reason to prefer it over Sonnet 5 or Opus 5.
Does Sonnet 5 have data retention?
No. Like Opus 5, Sonnet 5 does not require data retention. This is in contrast to Fable 5 and Mythos 5, which have 30-day retention policies. For sensitive codebases, Sonnet 5 is a safer choice than Fable.
What about the new tokenizer?
Sonnet 5 uses an updated tokenizer that can produce 1.0–1.35× more tokens for the same input text, depending on content type. The introductory pricing was set to make the transition roughly cost-neutral. After August 31, the standard pricing accounts for this.
Can I pin to a specific Sonnet 5 version?
Yes. Use the full model name claude-sonnet-5 instead of the sonnet alias. You can also set ANTHROPIC_DEFAULT_SONNET_MODEL in your environment to control which version the alias resolves to.
The Bottom Line
Claude Sonnet 5 is the model most coding agent users interact with every day. It delivers near-Opus performance at Sonnet prices, with a 1M context window, adaptive thinking, and no data retention requirements. The promotional pricing through August 31 makes this the most cost-effective time to use it for high-volume API work.
If you are on Pro or Team Standard, you already have it. If you are on Max and running Opus 5 for everything, consider switching to Sonnet 5 for routine tasks and saving Opus for the hard problems. The model is good enough that you will not notice the difference on most days.
Why pay $20/mo for ChatGPT, $20/mo for Claude, $20/mo for Gemini? aiFiesta gives you all of them and more for just $12/mo. One subscription, every answer.