Anthropic Claude Models in Borg

Claude models in Borg cover quick, bounded tasks as well as long coding and research work. Published gateway data gives measured response times where enough data is available.

Which Claude models are in Borg?

The Anthropic catalog lists six live models: Claude Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 4.5, Opus 4.6, and Sonnet 4.6. The catalog marks Opus 4.6 and Sonnet 4.6 as legacy models.

Model versions and tiers are not guarantees that an answer will be right. See the individual pages: Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, Claude Haiku 4.5, Claude Opus 4.6, and Claude Sonnet 4.6.

Choose by the work you need done, not the model name alone.

Choosing a Claude model

• Opus 5.5 is a starting point for long-running coding agents and involved knowledge work. Its adaptive thinking cannot be disabled.
• Sonnet 5.5 suits substantial, well-scoped work such as bug fixing, structured computer use, and creating documents, spreadsheets, slides, and design drafts.
• Haiku 4.5 is the compact, fastest Claude tier in the catalog. It fits quick code help, short extraction, and parallel classification or support work.
• Fable 5.1 is for demanding reasoning, long-horizon coding agents, multistep research, and creating documents, spreadsheets, and slides.

Haiku has smaller context and output limits than newer flagship models. Fable's adaptive thinking is always on, and the catalog suggests considering it when difficult work still fails at higher Opus effort. Opus 4.6 and Sonnet 4.6 remain available for workflows that rely on their behavior.

What the speed numbers mean

The figures quoted below use the Borg telemetry snapshot recorded at 2026-10-08T03:42:10.773437+00:00, covering a 30-day observation window from SearchAtlas AI Gateway production request logs. They are dated observations; the linked model pages show the current published measurements and each metric’s sample count.

Borg telemetry covers a 30-day window of production gateway request logs. Time to first response event measures how long the gateway takes to receive the first streamed response event. Completion speed is output tokens per second over the whole turn, including queueing and reasoning, not a per-token streaming rate.

Haiku 4.5 had 889 ms median time to first response event and 64.9 median completion tokens per second. Sonnet 5.5 measured 2,640 ms and 76.7 tokens per second; Opus 5.5 measured 2,946 ms and 60.0; Fable 5.1 measured 6,015 ms and 55.9. The 95th percentile for Haiku's first event was 2,307 ms.

These figures describe gateway observations, not a promise for every request.

The sample counts differ: Haiku had 6,197 first-event measurements, Sonnet 5.5 had 35,971, Opus 5.5 had 75,410, and Fable had 2,192. These are first-event sample counts, not the completion-speed denominators; check each model’s measurements for its metric-specific counts.

Success measurements and limits

The published success rate for Fable 5.1 is 99.9%. Other named Claude models have no published rate in this dataset because they are below the publication threshold. That does not mean they failed.

Telemetry has no coding-benchmark result for these Claude models. Missing data is not a score of zero. Start with the task, then compare answers on representative work. Borg Fusion can route work to suitable Claude lanes, or you can select a lane directly when a workflow needs a consistent model.

FAQ

Which Anthropic Claude models are included in Borg?

Borg includes Claude Fable 5.1, Opus 5.5, Sonnet 5.5, Haiku 4.5, Opus 4.6, and Sonnet 4.6.

Which Claude model is quickest in the published measurements?

Haiku 4.5 has a 889 ms median time to first response event in the published 30-day data.

Does Borg publish success rates for every Claude model?

No. A success rate is published for Fable 5.1, while other Claude success rates are below the publication threshold.