The Open-Source AI Stack
RSS
All models

Models · claude

Claude 3.5 Sonnet

Proprietary Anthropic · 2024-06-20 · Proprietary

The first Claude release to beat its own larger sibling (Claude 3 Opus) on most benchmarks. Established Artifacts, driving a wave of code-and-canvas product copies.

Cost

/ Mtok input
/ Mtok output

Anthropic API · as of 2026-05-21

source ↗

Speed

tok/sec output

Anthropic API · as of 2026-05-19

via Artificial Analysis ↗

Why people cared

Claude 3.5 Sonnet was the June 2024 release where a mid-tier model from a frontier lab decisively beat its own larger sibling. Claude 3 Opus from March had been the larger and more expensive offering; 3.5 Sonnet outperformed it across most benchmarks at one-fifth the price and twice the speed. The release also introduced Artifacts, a UI surface for code, diagrams, and document drafts rendered alongside the chat. Artifacts triggered a wave of product copies (canvas modes appeared in ChatGPT, Gemini, Mistral's chat product, and several open-weights wrappers within months) and established a new default expectation for what a chat interface was. On the API side, Claude 3.5 Sonnet held a leadership position on code benchmarks through most of late 2024 and became the default backend for early agentic coding tools (Cursor, Continue, Cline). The strategic story for Anthropic was that a smaller-faster-cheaper model from the mid tier could carry product surface area, which made Claude 3 Opus's slower follow-up effectively redundant; the model line skipped directly to 3.7 Sonnet in February 2025.

Architecture

tokens in Embedding vocab not disclosed × N layers Architecture not disclosed (proprietary or undocumented) Output projection tokens out
Schema-generated from data/models.yaml. Every label is auditable against the model's sources.

Specs

Architecture
unknown
Total params
not disclosed
Active params
not disclosed
Context window
200K tokens
Attention
unknown
Position encoding
unknown
Post-training
rlhf, constitutional
OSI-approved
no
Data released
no
Training code
not released

Benchmarks

Each score carries the date it was published; we never infer or interpolate missing scores.

Recommended use cases

  • general chat
  • code generation
  • tool use
  • Artifacts-style canvas interactions

Available quantizations

None. The weights are not distributed, so there are no public quantizations.

Notable innovations

  • · Mid-tier beat its own large sibling
  • · Artifacts product surface

Known limitations

  • · Architecture not disclosed; comparisons rest on published benchmarks only. source ↗

Lineage

Derivatives

Sources