Anthropic Claude (2026 line-up) — Architecture, Capabilities, Enterprise Positioning
As of 2026-10-06
What is Anthropic Claude (2026 line-up)?
As of 6 October 2026 Anthropic's current Claude line is Fable 5.1 (most capable), Opus 5.5 (recommended starting point), Sonnet 5.5 and Haiku 4.5, sold through the Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry and Claude Platform on AWS; SAP customers reach Claude through the generative AI hub, where the latest listed version is Opus 4.8 — so model choice starts with SAP Note 3437766, not with Anthropic's release notes.
The current line-up (checked 6 October 2026)
Anthropic's models overview lists four current models:
| Model | API ID | Positioning | Context / max output | Price per million tokens (input / output) |
|---|---|---|---|---|
| Claude Fable 5.1 | claude-fable-5-1 | Demanding reasoning and long-horizon agentic work | 1M / 128K | $10 / $50 |
| Claude Opus 5.5 | claude-opus-5-5 | Long-running agentic coding and knowledge work; Anthropic's recommended starting point | 1M / 128K | $4 / $20 |
| Claude Sonnet 5.5 | claude-sonnet-5-5 | Best combination of speed and intelligence | 1M / 128K | $2 / $10 |
| Claude Haiku 4.5 | claude-haiku-4-5-20251001 | Fastest model | 200K / 64K | $1 / $5 |
All current models accept text and image input, produce text, and support tool use and multilingual work. Fable 5.1 and Opus 5.5 use adaptive thinking that is always on; depth is steered with an effort parameter whose default is high on Fable 5.1 and Sonnet 5.5 and medium on Opus 5.5. Haiku 4.5 uses the earlier extended-thinking mode. Batch requests cost 50% less, and prompt-cache reads are billed at a fraction of the input price. Anthropic publishes a minimum retirement date for each model on its own platforms (for example not before 22 September 2027 for Opus 5.5 and not before 15 October 2026 for Haiku 4.5); Amazon Bedrock and Google Cloud set their own lifecycle dates.
Why it matters
- The current line (6 October 2026) is Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 4.5, with published prices, context windows and minimum retirement dates — a factual basis for tier and cost decisions.
- SAP's generative AI hub runs on a different clock: its 4 September 2026 What's New lists Claude up to Opus 4.8, flagged not for sovereign cloud offerings.
- Channel choice (Claude API, Bedrock, Vertex AI, Foundry, Claude Platform on AWS, or SAP) determines which model you can promise, under which contract and data terms.
Key points
- Current models: Claude Fable 5.1 ($10/$50 per M tokens), Opus 5.5 ($4/$20, recommended start), Sonnet 5 ($2/$10), Haiku 4.5 ($1/$5); 1M context except Haiku 4.5 (200K).
- Legacy but available: Fable 5, Opus 5.5, Opus 4.8, 4.7, 4.6, 4.5, Sonnet 4.6, 4.5; Claude Mythos is restricted to Project Glasswing (announced 7 April 2026).
- Thinking and cost levers: adaptive thinking with an effort parameter (default medium on Opus 5.5, high on Fable 5.1 and Sonnet 5), batch at 50% off, discounted prompt-cache reads.
- Channels: Claude API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Claude Platform on AWS — lifecycle dates on Bedrock and Vertex AI are set by those platforms.
- SAP: generative AI hub lists Anthropic models up to Opus 4.8 (July 2026) via Bedrock, not for sovereign offerings; Claude in Joule announced 12 May 2026 (C115).
- Retirement commitments are model-specific, not generation-wide: Anthropic-operated platforms commit to not retiring Opus 5.5 before 22 September 2027 and Haiku 4.5 before 15 October 2026 — build re-evaluation checkpoints around the fastest-expiring tier in a stack, not the newest one.
- Batch API and Message Batches: current models support up to 300K output tokens via the output-300k-2026-03-24 beta header on the Batch API, materially higher than the 128K synchronous cap — relevant for large offline document-review runs.
Terms used on this page
- Constitutional AI
- Anthropic's alignment approach: training the model against a written set of principles in addition to human feedback.
- Responsible Scaling Policy (RSP)
- Anthropic's published policy linking safety measures and deployment safeguards to model capability thresholds.
- Adaptive thinking and effort
- Current Claude models decide how much to reason per request; the effort parameter (low to max) steers depth, cost and latency.
- Claude Mythos
- Restricted Anthropic model line; Mythos Preview was announced on 7 April 2026 with Project Glasswing for defensive cybersecurity and is not generally available.
- SAP Note 3437766
- The SAP note listing the models and versions available in the generative AI hub of SAP AI Core.
Sources
- Anthropic — Claude models overview (checked 23 Sept 2026)
- Anthropic — Project Glasswing and Claude Mythos Preview (7 April 2026)
- SAP Help — SAP AI Core guide incl. What's New for the generative AI hub (PDF, 4 Sept 2026)
- SAP News — SAP and Anthropic: Claude on SAP Business AI Platform (12 May 2026)
- SAP News — SAP Unveils the Autonomous Enterprise (Sapphire 2026)
- AWS for SAP Blog — Joule for Consultants with Anthropic Claude models on Amazon Bedrock (11 Aug 2026)
- SAP Community — Building and consuming an MCP server in SAP Integration Suite, connecting it to Claude and Joule (11 Sept 2026)
- Anthropic — Model deprecations and retirement commitments
- Anthropic — Pricing (batch, prompt caching, per-model rates)
- Anthropic — Migrating to Claude Opus 5.5 guide
- Anthropic — Claude in Amazon Bedrock
- Anthropic — Claude on Vertex AI
- Anthropic — Claude Sonnet 5.5 overview (model ID, $2 / $10, 1M context, 128K output, platforms)
- Anthropic — Claude Sonnet 5.5 migration guide (breaking changes from Sonnet 5)
- Anthropic — Claude Platform release notes, 28 and 30 September 2026
- AWS Machine Learning Blog — Implementing Multi-Environment Access for Claude Platform on AWS (1 October 2026)
Full card available to members. What the full card adds: the full decision framework · the SAP vs Snowflake / Databricks / Fabric comparison · the common pitfalls and their fix · the cheat sheet · the architecture schemas · the code blocks · the facts worth quoting.