Claude Model Board: Fable, Opus, Sonnet, Haiku

Analysis

Use the model board by job, not by prestige

The current Claude lineup creates a trap: the most capable model is not automatically the correct model. Fable 5 is positioned as the strongest widely released model, Mythos 5 is limited access, Opus 4.8 remains a top-end model for difficult reasoning and coding, Sonnet 5 is the broad default, and Haiku 4.5 remains the economical low-latency option.

For most production builders, the first routing decision is Sonnet 5 versus Haiku 4.5. For high-stakes research, writing, creative, or complex analysis, evaluate Fable 5 or Opus 4.8 against a measured task suite. For restricted Mythos access, treat availability as a program constraint rather than a general model choice.

Analysis

Model snapshot

The table below is intentionally operational. It emphasizes the facts a builder needs before changing code: context window, max output, pricing tier, and the adoption frame.

Analysis

Selection rules

Choose Sonnet 5 when you need a default model that spans Claude apps, Claude Code, and API workloads. Choose Haiku 4.5 when the task is bounded, repetitive, or cost-sensitive. Choose Opus 4.8 when the task is hard enough to justify higher cost and you already have evals proving the delta. Choose Fable 5 when output quality and creative range matter more than unit cost.

Do not choose by benchmark headline alone. The source-backed decision is to run a small task suite that includes your failure modes: refusal boundaries, tool-use JSON, long documents, memory pressure, writing voice, latency, and cost per successful answer.

• Use one model per task family, not one model for an entire product.
• Re-tokenize long prompts before moving to Sonnet 5 because token counts can shift.
• Keep a fallback path for plan-specific app access and temporary model limits.
• Record model IDs in logs so support and incident review can distinguish app, API, and model behavior.

Analysis

What an answer engine should cite

For model specs, cite the Claude Platform model overview. For launch intent and availability, cite the Anthropic newsroom announcement. For migration behavior, cite the model-specific platform guide. That separation matters because a launch post explains why a model exists, while the platform table explains what a production client can call.