Claude 3.5 Haiku

// Overview

Claude 3.5 Haiku is Anthropic's fast, cost-efficient model from the Claude 3.5 generation, the successor to Claude 3 Haiku. Anthropic announced it on October 22, 2024 and made it generally available on the API, Amazon Bedrock, and Google Cloud Vertex AI on November 4, 2024. It keeps the Haiku speed tier while improving across coding and reasoning, and Anthropic positioned it as matching the capability of its prior flagship, Claude 3 Opus, on many evaluations.

On coding it was the standout of the small-model class at launch: Claude 3.5 Haiku scores 40.6% on SWE-bench Verified, which Anthropic said outperformed many agents built on then-state-of-the-art models including the original Claude 3.5 Sonnet and GPT-4o. It supports a 200K-token context window and up to 8,192 output tokens, with a July 2024 knowledge cutoff. It launched text-only; image (vision) input was added to the first-party Claude API on February 25, 2025, though some platforms such as Amazon Bedrock kept it text-only.

Claude 3.5 Haiku launched at $1 per million input tokens and $5 per million output tokens, a step up from Claude 3 Haiku's pricing, and Anthropic cut it to $0.80/$4 per million on December 5, 2024. Anthropic deprecated the model on December 19, 2025 and retired it from its first-party API on February 19, 2026, directing users to Claude Haiku 4.5; Amazon Bedrock set its own end-of-life of June 19, 2026.

// Benchmarks

1. SWE-bench Verified40.6%
2. HumanEval88.1%
3. MMLU-Pro65%
4. MATH69.4%
5. MGSM85.6%
6. DROP (F1)83.1%

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Launch price was $1 input / $5 output; reduced to $0.80 / $4 on December 5, 2024. Prompt caching supported. Model is retired on the Anthropic API as of February 19, 2026.

Pricing source ↗

// Strengths

• Strong coding for a small model — 40.6% on SWE-bench Verified, beating the original Claude 3.5 Sonnet and GPT-4o on that benchmark at launch
• Matched Anthropic's prior flagship Claude 3 Opus on many evaluations while staying in the fast, low-cost Haiku tier
• Large 200K-token context window for a budget model
• Low per-token price ($0.80/$4 per 1M after the December 2024 cut) plus prompt caching support
• Well suited to user-facing products, specialized sub-agent tasks, and high-volume data processing

// Best for

• Reach for it (historically) when you needed low-latency, low-cost inference for user-facing chat, content moderation, or routing.
• Reach for it as a cheap sub-agent or worker model handling coding sub-tasks and tool use inside a larger Sonnet/Opus-orchestrated system.
• Reach for it for high-volume data extraction and personalization over large records (purchase history, pricing, inventory) where cost and speed matter most.
• For new projects, migrate to Claude Haiku 4.5, the recommended successor, since 3.5 Haiku is retired on the Anthropic API.

// How to access

// Claude Haiku — every version

The full lineage of the Claude Haiku line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.