Haiku
The lightweight, fast model tier in Anthropic's Claude series, used for high-frequency, low-cost calls such as summarization, file naming and command completion.
Core Facts
Timeline (last 90 days)
The author suggests: if there is a real use case, start with APIs such as Haiku; self-hosting (renting GPUs) is only justified under hard constraints such as compliance, privacy, or offline deployment.
The cost of the Claude API is primarily determined by token consumption, including input context length, output length, and model version, with price differences between Opus/Sonnet/Haiku reaching several times over.
The team overrode the Haiku slot with DeepSeek V4 Flash through local configuration, having it handle templated scripting tasks such as Remotion video scripts, subtitle exports, and screenshot packaging.
Anthropic's Claude model family is tiered by capability: Opus is the flagship tier suited for complex reasoning and is the most expensive, Sonnet is the mid-tier balanced option, and Haiku is the lightweight, fast, low-cost option.
In Cowork's model selection, Haiku is the cheapest, Sonnet is mid-range, and Opus is the most capable but consumes the most tokens
Among Claude model options, Opus 4.6 is suited for heavy tasks but is unavailable in the free tier, Sonnet 4.6 is the default model for the free tier, and Haiku is faster and consumes fewer tokens.
The early Haiku system insisted on developing its own toolchain and suffered from a long-term shortage of software.
Haiku is the smallest, lightest, and fastest model in the Claude series, suitable for large-scale batch processing and tasks requiring rapid response.
A Skill can individually specify the model to use, for example being fixed to use Sonnet or Haiku to save costs.
Claude Code's Explorer agent previously defaulted to the lightweight Haiku model, but was later changed to match the main model — if Opus is used, Explorer also runs on Opus, making searches slower and more expensive.
40 more timeline events
All Facts (20)
In Anthropic's model naming system, Opus represents the most powerful version, Sonnet is the mid-tier version, and Haiku is the lightweight version.
65%UnverifiedAnthropic's Claude model naming system borrows from musical terminology: Haiku (lightweight and fast), Sonnet (balancing performance and speed), Opus (highest performance)
90%UnverifiedAnthropic推荐使用编排器+子代理架构，用推理能力强的模型负责规划决策，用Sonnet或Haiku执行具体点击操作
90%Unverified在Anthropic的模型体系中，Opus系列定位为最高能力层级的模型，在推理深度、代码生成质量和长上下文理解方面优于Sonnet和Haiku系列
90%UnverifiedIn Anthropic's model naming system, Opus, Sonnet, and Haiku correspond to high, medium, and low tiers of capability and cost respectively.
60%UnverifiedThe author suggests: if there is a real use case, start with APIs such as Haiku; self-hosting (renting GPUs) is only justified under hard constraints such as compliance, privacy, or offline deployment.
50%UnverifiedThe cost of the Claude API is primarily determined by token consumption, including input context length, output length, and model version, with price differences between Opus/Sonnet/Haiku reaching several times over.
50%UnverifiedThe team overrode the Haiku slot with DeepSeek V4 Flash through local configuration, having it handle templated scripting tasks such as Remotion video scripts, subtitle exports, and screenshot packaging.
50%UnverifiedAnthropic's Claude model family is tiered by capability: Opus is the flagship tier suited for complex reasoning and is the most expensive, Sonnet is the mid-tier balanced option, and Haiku is the lightweight, fast, low-cost option.
50%UnverifiedIn Cowork's model selection, Haiku is the cheapest, Sonnet is mid-range, and Opus is the most capable but consumes the most tokens
50%UnverifiedAmong Claude model options, Opus 4.6 is suited for heavy tasks but is unavailable in the free tier, Sonnet 4.6 is the default model for the free tier, and Haiku is faster and consumes fewer tokens.
50%UnverifiedThe early Haiku system insisted on developing its own toolchain and suffered from a long-term shortage of software.
50%UnverifiedHaiku is the smallest, lightest, and fastest model in the Claude series, suitable for large-scale batch processing and tasks requiring rapid response.
50%UnverifiedA Skill can individually specify the model to use, for example being fixed to use Sonnet or Haiku to save costs.
50%UnverifiedClaude Code's Explorer agent previously defaulted to the lightweight Haiku model, but was later changed to match the main model — if Opus is used, Explorer also runs on Opus, making searches slower and more expensive.
50%UnverifiedThe Explore agent uses the Haiku model, because the task of searching for content within a codebase does not require reasoning capabilities.
50%UnverifiedCode review tasks are adequately handled by Sonnet or even Haiku, and should not be run with Opus (with effort max enabled).
50%UnverifiedA fixed QA process can be encapsulated as a Skill and executed with the lighter-weight Haiku model to save costs.
50%UnverifiedHaiku is the lowest-cost, fastest version in the Anthropic model family, suitable for handling structured and highly repetitive tasks.
50%UnverifiedIn a demonstration classifying 100 emails (hot leads/cold leads/non-leads), Jev completed all classifications in under one second, outperforming Haiku and Fable in both speed and cost.
50%