Verified July 27, 2026
Claude Fable 5 integrations
Every tool below was checked against its own documentation, changelog or issue tracker — including the ones where the honest answer is that it does not work yet.
How to read this page
Three patterns cover almost everything here. Tools that resolve models from a catalog need the catalog to know about claude-fable-5, and break in confusing ways when it does not. Tools with a config file usually accept the ID immediately, then fail on a parameter this model removed — temperature, or a thinking budget. Tools that wrap another agent inherit that agent's model list, which is where the worst failures live, because the error surfaces far from its cause.
Cursor — Fable 5.1 on launch day
claude-fable-5-1 landed in Cursor with the September release and sits at the top routing tier. See our Cursor guide for Max Mode billing, bundled quota, BYOK, and the model-badge check.
OpenRouter — one key, watch the markup
Route anthropic/claude-fable-5-1 through OpenRouter for unified billing across providers. Credit markup, cache_control passthrough, and refusal handling are on you — not automatic.
Vertex AI — GCP enterprise front door
Model Garden serves claude-fable-5-1 for teams already on Google Cloud. Region filters, zero default quota, and 30-day retention still apply.
Claude Code
Anthropic's official coding agent — select claude-fable-5 in /model, set effort, and verify it isn't silently falling back to Opus 4.8.
Zed — works via BYOK, blocked over ACP
Native Anthropic BYOK has shipped Fable 5 since June 2026, behind a data-retention consent setting. The Claude Agent path validated the model ID at the ACP bridge and threw for claude-fable-5 until claude-agent-acp 0.55.0.
Cline — supported, cost is the real question
Available through both Cline's usage billing and the Anthropic provider. The model list is curated, so custom IDs need a LiteLLM or OpenAI-compatible route. Split Plan and Act across two models before you do anything else.
Windsurf — supported, effort is a model entry
Cascade lists Fable 5 at five effort tiers as five separate picker entries, priced 40 to 100 credits on legacy credit plans. Free-tier accounts cannot reach the model at all.
Continue.dev — runs, but effort is not exposed
Any model ID works in config.yaml, so chat, edit and apply are fine. The reasoning options target the legacy thinking budget and return a 400, and we found no effort passthrough — so every request runs at the default high.
Aider — supported, feed it the metadata
aider --model claude-fable-5 runs immediately. Without a metadata file Aider assumes an unlimited context window and zero cost, and its temperature default plus the thinking-tokens switch both produce 400s.
OpenCode — supported on 1.17.0+, don't override
One config line sets it as the default. Overriding the catalog entry puts you on the generic-Anthropic path, which skips Fable reasoning handling and returns nothing at all — zero tokens, no error, no log line.
LangChain / LangGraph — full support both languages
effort and task budgets live under output_config in Python; the JavaScript package added Claude 5-series handling in June 2026. The recurring failure is a temperature argument inherited from an older chain.
Vercel AI SDK — supported directly and via Gateway
The provider package added claude-fable-5 alongside the fallbacks parameter, and AI Gateway serves the same model string with cross-provider failover. Plan for refusals: they arrive as successful responses.
Kiro — see the dedicated guide
Kiro's spec-driven workflow changes how a long-horizon model is worth using, so its setup and cost picture differ from the editors above. We cover it separately.
Not sure which of these is actually broken right now?
Model availability and client support both moved several times since launch. The status page records what we last confirmed, tool by tool, with the date attached.