Claude Fable 5.5 — release date, what is confirmed, and the Fable API today

Claude Fable 5.5 has not been released. As of October 2, 2026, Anthropic has published no model ID, price, model page or system card for it. The newest Fable is Claude Fable 5.1 (claude-fable-5-1), released September 1, 2026 at $10.00 input / $50.00 output per 1M tokens. The Claude 5.5 family so far is Opus 5.5 (September 22, 2026) and Sonnet 5.5 (September 28, 2026), and Anthropic's announcement names Haiku 5.5 as the next one, not a Fable. On Kunavo, Fable 5.1 costs $7.00 / $35.00 per 1M tokens today, 30% under Anthropic's list, and the model string is the only thing that changes when a newer Fable arrives.

Is Claude Fable 5.5 out? What is confirmed

Checked against Anthropic's own pages on October 2, 2026:

The status above is re-checked on this page; the dated log at the end records each change.

When could Fable 5.5 come out?

Anthropic's release record for the Fable line and the 5.5 family is short, and it is the only evidence there is:

One gap of 84 days between Fables is a single interval, not a schedule, and Anthropic has not said whether the 5.5 family will include a Fable at all. What is fixed: a new Fable would not take Fable 5.1 away — Anthropic commits to keep claude-fable-5-1 on its platform until at least September 1, 2027.

What Fable 5.5 might cost

Unknown until Anthropic publishes it. The pattern so far: Fable 5 and Fable 5.1 both listed at $10.00 / $50.00 per 1M tokens. Of the two 5.5 models released, Opus 5.5 cut 20% off Opus 5's list ($5.00 / $25.00 to $4.00 / $20.00) and Sonnet 5.5 kept Sonnet 5's $2.00 / $10.00. Kunavo sets a price when it adds a model; today every Claude model here sells below Anthropic's list, Fable 5 and 5.1 both at $7.00 / $35.00.

Fable 5.1 or Opus 5.5 today?

The question the "Fable 5.5" searches keep landing on is what Fable is for now that Opus 5.5 exists. Anthropic's own answer: start with Opus 5.5 for most workloads, and use Fable 5.1 "for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short." On the three benchmarks Anthropic published for both, Opus 5.5 scores higher (Anthropic's figures, not Kunavo's):

Anthropic rates Fable 5.1's latency "slower" and Opus 5.5's "moderate"; Fable 5.1 defaults to high effort and Opus 5.5 to medium. Both think on every request, both have a 1M-token context window and 128K output. The prices, per 1M tokens:

On one coding-agent step — 25K tokens in, 1.2K out — Fable 5.1 costs $0.217 on Kunavo and Opus 5.5 $0.087, 2.5× apart. Caching barely moves the ratio: with 90% of the input read from cache, as in a long agent session, the same step costs $0.063 against $0.027, still 2.4×, because output — thinking included — is most of a cached step on both models. Claude Sonnet 5 at $1.40 / $7.00 is cheaper again; every Claude rate is in the Claude API pricing guide.

Be ready for Fable 5.5 on the day it ships

One Kunavo key reaches every Fable it serves through the OpenAI-compatible endpoint and the native Anthropic Messages API. Keep the model string in configuration and the switch is one variable:

import os
from openai import OpenAI

client = OpenAI(
    api_key=os.environ["KUNAVO_API_KEY"],
    base_url="https://api.kunavo.com/v1",
)

# Today: claude-fable-5-1. The day a Fable 5.5 is listed, change the variable.
MODEL = os.environ.get("CLAUDE_MODEL", "claude-fable-5-1")

resp = client.chat.completions.create(
    model=MODEL,
    max_tokens=16000,   # Fable always thinks; thinking bills as output
    messages=[{"role": "user", "content": "Plan the migration, then list the risks."}],
)
print(resp.choices[0].message.content)

To see which Fable IDs your key can call right now, list the models:

curl -s https://api.kunavo.com/v1/models \
  -H "Authorization: Bearer $KUNAVO_API_KEY" | grep -o '"claude-fable[^"]*"'

Three habits make the switch a non-event:

• Do not send sampling parameters. Fable 5 and 5.1 answer temperature, top_p and top_k with a 400; Kunavo strips them from claude-fable-5-1 requests before they leave the gateway, but code that never sends them works on any model.
• Budget output for thinking. Fable 5.1 thinks on every request at high effort by default, and thinking bills as output. Size max_tokens for thinking plus the answer; do not hard-code budget_tokens, which the newest models reject.
• Replay thinking blocks unchanged. They are bound to the model and conversation; switch models at the start of a conversation, not in the middle — the thinking-signature error guide covers the error otherwise.

For Claude Code, set the session model and pin the aliases:

export ANTHROPIC_BASE_URL=https://api.kunavo.com   # origin, no /v1
export ANTHROPIC_AUTH_TOKEN=sk-kn-...

# Run the session on the newest Fable Kunavo serves.
export ANTHROPIC_MODEL=claude-fable-5-1
# Pin the aliases to models Kunavo serves. The sonnet alias asks for
# Sonnet 5.5, which Kunavo does not serve: without the SONNET line,
# /model sonnet, opusplan's execution phase and sonnet subagents get a 404.
# Opus 5.5 needs Claude Code v2.1.280 or later (claude update).
export ANTHROPIC_DEFAULT_OPUS_MODEL=claude-opus-5-5
export ANTHROPIC_DEFAULT_SONNET_MODEL=claude-sonnet-5
export ANTHROPIC_DEFAULT_HAIKU_MODEL=claude-haiku-4-5

claude

The rest of the setup is on the Claude Code page; keys are created at /app/keys.

Status log

• October 2, 2026 — No Claude Fable 5.5. Anthropic's models overview lists Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5; its newsroom has no Fable post after September 1. Kunavo serves Fable 5.1, Fable 5 and Opus 5.5.

FAQ