Claude Haiku 4.5
Claude Haiku 4.5 by Anthropic adds extended thinking, context awareness, and computer use to the Haiku line for sub-agent roles and agentic workflows.
Model Overview
Capabilities, design details, and architectural traits
Claude Haiku 4.5 - First Haiku with Extended Thinking
Claude Haiku 4.5 is a model in Anthropic's 4.5 family. It is the first model in the Haiku line to support extended thinking, context awareness, and computer use - three capabilities previously exclusive to Sonnet and Opus tiers. This makes it a structurally different model from all prior Haiku releases, not just a faster version of the same design.
| Differentiator | Detail |
|---|---|
| Extended thinking - first for Haiku tier | Supports interleaved thinking between tool calls, summarized thinking output, and configurable thinking token budgets; prior Haiku models had no extended thinking mode |
| Context awareness | Receives real-time token budget updates after each tool call, enabling the model to track remaining context window and avoid premature task abandonment during long-running agent sessions |
| Computer use | Supported for the first time in the Haiku line; Haiku 3.5 did not support computer use |
| Documented sub-agent role | Officially positioned for multi-agent systems where a Sonnet model orchestrates and multiple Haiku 4.5 instances execute parallel subtasks |
| ASL-2 classification | Released under AI Safety Level 2 - lower than Sonnet 4.5 and Opus 4.1 (ASL-3), reflecting different CBRN risk profile |
| Drop-in replacement scope | Serves as a direct replacement for both Haiku 3.5 and Sonnet 4 |
Capability Tier Shift Within the Haiku Line
Haiku 4.5 is not an incremental update to Haiku 3.5 - it introduces a new capability baseline for the Haiku class. Extended thinking, context awareness, and computer use each represent features Haiku 3.5 entirely lacked. The combination changes which workloads the Haiku tier can handle without escalating to a larger model, particularly for parallel agent execution and real-time computer-driven tasks.
Benchmark Performance
Independent evaluations · Artificial Analysis
Accuracy & Capability Details
Compare Models Side-by-Side
Evaluate specifications, pricing, and independent benchmark indices
| Model Details | |||
|---|---|---|---|
| General Info | |||
| Provider | Anthropic | Anthropic | Anthropic |
| Release Date | October 15, 2025 | September 22, 2026 | September 28, 2026 |
| Knowledge Cutoff | - | - | Jun 2026 |
| Context & Limits | |||
| Context Window | 200K | 1M Best Context Window | 1M Best Context Window |
| Pricing (per 1M tokens) | |||
| Input Pricing | $1 Best Input Pricing | $4 | $2 |
| Output Pricing | $5 Best Output Pricing | $20 | $10 |
| Modalities | |||
| Inputs | textimagefile | textimagefile | textimagefile |
| Outputs | text | text | text |
| Benchmarks (0-100) | |||
| Intelligence Index | 15.4 | 57.6 Best Intelligence Index | 56.0 |
| Coding Index | - | - | - |
| Agentic Index | - | - | - |
Humanity's Last Exam
Extremely difficult logical reasoning and knowledge.
Long Context Reasoning
Logical reasoning over long context windows.
SciCode Benchmark
Scientific coding and mathematical modeling.
Independent evaluation data provided by Artificial Analysis. To view the latest benchmarks and full details, visit their official site.
Explore more from Anthropic
Other models by Anthropic
Claude Opus 5.5
Claude Sonnet 5.5
Claude Fable 5.1
Claude Opus 5
Top AI Models
Leading alternatives by intelligence score
