Claude Sonnet Versions Compared: Every Release, Specs, Pricing and Changes

Claude Sonnet is Anthropic's mid-tier Claude model line, whose first version, Claude 3 Sonnet, carries a February 2024 snapshot date, and nine Sonnet versions have appeared in Anthropic's documentation since then. This page compares every Sonnet version side by side: API IDs, release dates, status, context window, maximum output, prices, documented behaviour changes, published benchmark figures and retirement dates. It is written for developers migrating between versions, buyers estimating cost and architects who need to know which Sonnet ID is still callable. Versions are listed by release date, newest first, an objective order that implies no ranking. Every fact comes from Anthropic's own documentation and announcements, cited in the sources list.

Last checked: 2026-11-15.

This is an independent reference page from NHI Mgmt Group based on public provider documentation; it is not sponsored or endorsed by any provider. It uses Anthropic's public documentation as of November 2026. Model details change, so confirm against Anthropic's current documentation.

Key takeaways

• Three Sonnet versions are active. Anthropic's deprecations page lists claude-sonnet-5-5, claude-sonnet-5 and claude-sonnet-4-6 as Active; Sonnet 4.5 is deprecated and every earlier Sonnet is retired.
• Sonnet 4.5 retires on 30 November 2026. Anthropic announced the deprecation of claude-sonnet-4-5-20250929 on 30 September 2026 and names claude-sonnet-5-5 as the replacement.
• Prices fell from $3/$15 to $2/$10. Sonnet 3.7 to 4.6 list $3 input and $15 output per million tokens; Sonnet 5 and 5.5 list $2 and $10, and Sonnet 5.5's cache read is $0.10.
• Context grew from 200K to 1M tokens. Sonnet 4 and 4.5 list 200K tokens and 64K output; Sonnet 4.6, 5 and 5.5 list 1M tokens and 128K output at standard pricing.
• Version upgrades can break code. Sonnet 5 rejects manual extended thinking and non-default sampling parameters, and Anthropic lists five ways Sonnet 5 code can break on Sonnet 5.5.

What is Claude Sonnet?

Claude Sonnet is one of five Claude model lines in Anthropic's documentation, alongside Fable, Mythos, Opus and Haiku. Anthropic's models overview rates Sonnet 5.5 latency as Fast and prices it below Opus 5.5 and above Haiku 5.5. Every Sonnet model is proprietary: Anthropic publishes no weights and states no parameter count for any version. Sonnet 4.6 and later use undated API identifiers such as claude-sonnet-5-5; earlier versions carry a date snapshot, such as claude-sonnet-4-5-20250929.

Anthropic's numbering is not continuous across the line. There was no Sonnet 4.1 and no Sonnet 4.7 or 4.8; the line moved from Sonnet 4.6 (February 2026) directly to Sonnet 5 (June 2026). Claude 3.5 Sonnet was released twice under the same name, as claude-3-5-sonnet-20240620 and the upgraded claude-3-5-sonnet-20241022. For the wider Claude line-up, see the Anthropic Models hub.

Every model at a glance

As of 2026-11-15. Source: Anthropic models overview, model deprecations page and release notes.

All Sonnet versions take text and image input and return text, per Anthropic's models overview for the current models. Open weights: none for any version. Licence: proprietary. Parameters: not stated by Anthropic.

When was each model released?

As of 2026-11-15. Dates are as shown in Anthropic's Claude Platform release notes.

How much does each model cost?

As of 2026-11-15. USD per million tokens, from Anthropic's pricing page. Claude Sonnet 4 is still listed there at $3/$15 although retired; retired versions are omitted here.

Tokeniser boundary: Anthropic's pricing page states that Claude 4.7 and later models use a newer tokeniser producing about 30% more tokens for the same text, so Sonnet 5 and 5.5 per-token prices are not directly comparable with Sonnet 4.6 and earlier. Anthropic's Sonnet 5 announcement gives the range as roughly 1.0 to 1.35 times, depending on content type. Re-measure token counts on your own prompts before comparing cost.

Data residency: For Sonnet 4.6 and later, setting inference_geo to "us" applies a 1.1x multiplier to all token categories; earlier models return a 400 error for that parameter. Fast mode is not offered for any Sonnet version on the pricing page. For cost guardrails, see How to set spend limits on LLM API keys.

What changed between models?

Only changes Anthropic documents in its release notes and announcements are listed.

• Sonnet 3.5 to 3.7: Sonnet 3.7 added a choice between near-instant responses and step-by-step extended thinking. The upgraded Sonnet 3.5 (October 2024) had introduced computer use tools.
• Sonnet 3.7 to 4: Sonnet 4 launched with extended thinking alongside Opus 4 and later gained a 1M-token context beta (August 2025).
• Sonnet 4 to 4.5: Anthropic's announcement keeps pricing the same as Sonnet 4, adds context editing and a memory tool on the API, and states the model ships under AI Safety Level 3 (ASL-3) protections. Structured outputs reached general availability on Sonnet 4.5 on 2026-01-29.
• Sonnet 4.5 to 4.6: Sonnet 4.6 supports adaptive and extended thinking and context compaction (beta), became the default model for Free and Pro users on claude.ai, and moved to a 1M window at standard pricing in March 2026. From 2026-11-03, Sonnet 4.6 text carries Anthropic's text watermark; with non-default temperature, top_p or top_k the watermark might be absent. See the AI Model Provenance and Watermarking Guide.
• Sonnet 4.6 to 5: adaptive thinking is on by default, manual extended thinking returns a 400 error, and non-default temperature, top_p and top_k return a 400 error. The newer tokeniser applies. Anthropic states cyber safeguards are enabled by default and reports lower hallucination and sycophancy rates than Sonnet 4.6. Computer use and browser use toolsets (computer_toolset_20260801, browser_toolset_20260801) became available for Sonnet 5 on the Claude API on 2026-08-19.
• Sonnet 5 to 5.5: Anthropic lists five ways Sonnet 5 code can break, including forced tool use (tool_choice types any and tool) returning a 400 error, and the older computer_20251124 tool is not accepted on the Claude API or Google Cloud. The announcement states output generation is over 30% faster than Sonnet 5, adds a between_tools thinking setting, and states that Sonnet 5.5 includes cyber safeguards similar to Opus 5.5, with higher-risk cyber requests falling back to Sonnet 5, and adds classifiers that prevent reasoning extraction. The models overview lists adaptive thinking with a default effort of high.

For background on thinking modes and effort settings, see Reasoning Models Explained.

What workloads does Anthropic position each model for?

This section records the provider's own positioning wording; it is not an assessment.

Worked cost example

Using the list prices above, a workload of 10,000 requests a day, each with a 20K-token prompt and a 1K-token response, uses 200 million input and 10 million output tokens a day. On Sonnet 4.6 or 4.5 that is $600 input plus $150 output, $750 a day. On Sonnet 5 or 5.5 it is $400 input plus $100 output, $500 a day, before the tokeniser difference; at the upper end of Anthropic's stated 1.0 to 1.35 times range, the same text would cost about $675 a day. Prompt caching changes the result most: Sonnet 5.5's $0.10 cache read is half of Sonnet 5's. Re-run the arithmetic with token counts measured on your own data.

What is the context window of each model?

As of 2026-11-15. Context, maximum output and knowledge cutoff from Anthropic's models overview, deprecations page and release notes; reasoning controls as documented in the release notes, announcements and models overview.

Sonnet 4.6, 5 and 5.5 list a 1M-token context window and 128K maximum output, and Anthropic's pricing page states the full window is billed at standard rates, so a 900K-token request costs the same per token as a 9K-token request. Sonnet 4.5 and Sonnet 4 list 200K tokens and 64K output. Both previously had a 1M-token beta (header context-1m-2025-08-07) that Anthropic retired on 2026-04-30; since then, requests over 200K tokens to those models return an error. Sonnet 3.5 supported 8,192 output tokens from August 2024. Context windows for Sonnet 3, 3.5 and 3.7 are not stated on the current pages checked.

Capabilities and tools compared

As of 2026-11-15. Anthropic's models overview states that "all current models support text and image input, text output, multilingual capabilities, vision, and tool use". Cloud IDs are from the individual model overview pages on platform.claude.com.

Fine-tuning is not mentioned for any Sonnet version on the Anthropic pages checked; partner clouds may publish their own options, so confirm in each cloud's catalogue.

Where is each model available?

Partner clouds publish their own model IDs and retirement schedules, which can differ from the Claude API. Confirm in each cloud's model catalogue.

Model cards and system cards

System card links are as given in each announcement; this page does not summarise their contents.

Benchmarks

As of 2026-11-15. Figures are provider-reported, as stated in the text of each Anthropic announcement. Benchmark suites and versions differ between announcements, so figures in different rows are not directly comparable.

Deprecations and retirements

As of 2026-11-15. Source: Anthropic model deprecations page. Anthropic states it gives at least 60 days' notice before retiring publicly released models.

Requests to a retired ID return an error. For lifecycle planning, see the AI Model Retirement Guide and the LLM Deprecation and Retirement Tracker.

Data handling and security

Anthropic's zero data retention page designates only the Fable and Mythos 5.x models as Covered Models requiring 30-day retention; no Sonnet version is on that list, and the Sonnet 5.5 announcement states zero data retention is supported. Regardless of arrangement, the page states flagged content may be retained for up to 2 years. Confirm retention terms in your own agreement. Sonnet 5 and 5.5 enable cyber safeguards by default, which can refuse or route some security-related requests.

The main operational risk across Sonnet versions is the credential, not the model: Claude API keys grant access to every model the workspace can call. See the Anthropic API Keys Security Guide, the LLMjacking Guide and, for coding agents built on Sonnet, the Claude Code Security Guide.

How to choose a model: factual criteria

Run your own evaluation on representative prompts before switching versions; provider benchmarks use different suites per release.

Practitioner checklist

• Inventory every service, agent and script still calling claude-sonnet-4-5-20250929 and migrate before 30 November 2026.
• Pin production code to an explicit Sonnet API ID and record it in the AI inventory.
• Test Sonnet 5 and 5.5 migrations for 400 errors from manual extended thinking, sampling parameters and forced tool use.
• Re-measure token counts on real prompts when crossing from Sonnet 4.6 to Sonnet 5 or later.
• Check partner cloud model catalogues separately for their own retirement dates.
• Confirm data retention terms for the chosen version against your agreement.
• Store Claude API keys in a secrets manager and set workspace spend limits.

Standards and references

• Anthropic: Model deprecations (accessed 2026-11-15)
• Anthropic: Models overview (accessed 2026-11-15)
• Anthropic: Pricing (accessed 2026-11-15)
• Anthropic: Claude Platform release notes (accessed 2026-11-15)
• Anthropic: Zero data retention (accessed 2026-11-15)
• Anthropic: Introducing Claude Sonnet 5.5 (accessed 2026-11-15)
• Anthropic: Introducing Claude Sonnet 5 (accessed 2026-11-15)
• Anthropic: Introducing Claude Sonnet 4.6 (accessed 2026-11-15)
• Anthropic: Introducing Claude Sonnet 4.5 (accessed 2026-11-15)

Related NHI Mgmt Group resources: Anthropic Models, Claude Haiku Versions Compared, Frontier AI Models Compared, LLM Pricing Tracker, Claude for Enterprise Security Guide.