Premium museum expansion · Provider catalog
Anthropic
Claude — assistant and API models built with Constitutional AI and safety research narratives. Tiers Opus (depth), Sonnet (balanced), Haiku (speed) persist across generations. Current frontier SKUs ship on the Claude API, AWS Bedrock, and Google Vertex AI with parallel model IDs. See Anthropic and Claude (language model), March 2026 snapshot.
Architecture & training (overview)
Claude models are decoder-only transformers with long-context design (100K+ token class in Claude 2 era; later 200K+ and 1M-class claims on select SKUs). Training emphasizes RLHF-style alignment and Anthropic’s Constitutional AI: a written set of principles used in supervised revision rounds and RLAIF (reinforcement learning from AI feedback) so preferences can be steered without relying only on human labels — exact data mixes and parameter counts for frontier weights remain undisclosed.
Official: anthropic.com · Claude API docs · Models overview.
Tiers & naming
For complex coding and long-horizon agents, product docs often recommend starting from the strongest tier (Opus) and scaling down for latency or cost. Current Claude SKUs commonly accept text and image input and return text; multilingual use and vision are standard across the frontier line — always confirm per model card.
- Opus — maximum capability / coding depth; flagship for agents.
- Sonnet — daily-driver balance; Thinking variants in IDEs add reasoning budget.
- Haiku — fastest / lowest cost; summaries and high-volume tasks.
- Generations — Claude 3, 3.5, 4, 4.5, 4.6 denote major capability jumps in product naming.
Launch & benchmarks (selected)
Claude 3 (March 2024) introduced Opus / Sonnet / Haiku with strong multimodal benchmarks. Claude 3.5 Sonnet (mid-2024) often matched or exceeded prior Opus-class tasks at lower cost. Claude 4 and 4.5 (2025) added agentic tooling narratives (e.g. code execution, connectors, files API in industry coverage). 4.6 (2026) continues the same tier names with heavier IDE and long-task framing — verify scores on Anthropic’s announcements and third-party leaderboards; numbers move every quarter.
Frontier lineup — Opus 4.6, Sonnet 4.6, Haiku 4.5
Side-by-side snapshot of the latest three-tier stack: roles, API strings, major cloud IDs, list pricing, and reasoning flags. IDs and prices change — use this as a museum bookmark, not a billing source.
| Feature | Claude Opus 4.6 | Claude Sonnet 4.6 | Claude Haiku 4.5 |
|---|---|---|---|
| Role | Most capable for agents & coding | Best mix of speed & intelligence | Fastest; near-frontier quality |
| Claude API model ID | claude-opus-4-6 |
claude-sonnet-4-6 |
claude-haiku-4-5-20251001 |
| API alias (when applicable) | claude-opus-4-6 |
claude-sonnet-4-6 |
claude-haiku-4-5 |
| AWS Bedrock ID | anthropic.claude-opus-4-6-v1 |
anthropic.claude-sonnet-4-6 |
anthropic.claude-haiku-4-5-20251001-v1:0 |
| Vertex AI model ID | claude-opus-4-6 |
claude-sonnet-4-6 |
claude-haiku-4-5@20251001 |
| List pricing (API, USD / MTok) | $5 in · $25 out | $3 in · $15 out | $1 in · $5 out |
| Extended thinking | Yes | Yes | Yes |
| Adaptive thinking | Yes | Yes | No |
Pricing is the public list tier from provider docs (per million tokens). Discounts, batch, and priority classes are omitted. Anthropic pricing
Platform snapshot (Vertex-style fields)
Operational differentiators often called out next to cloud consoles: latency band, context/output ceilings, and stated cutoffs. Values vary by region and SKU — confirm in-console.
| Field | Opus 4.6 | Sonnet 4.6 | Haiku 4.5 |
|---|---|---|---|
| Priority tier eligible | Yes | Yes | Yes |
| Comparative latency | Moderate | Fast | Fastest |
| Context window (advertised) | Up to ~1M tokens | Up to ~1M tokens | ~200K tokens |
| Max output (advertised) | ~128K tokens | ~64K tokens | ~64K tokens |
| Reliable knowledge cutoff (stated) | May 2025 | Aug 2025 | Feb 2025 |
| Training data cutoff (stated) | Aug 2025 | Jan 2026 | Jul 2025 |
Model versions (consumer / API line)
High-level lifecycle labels — use the frontier and release tables above for exact version strings.
| Name | Era | Status | Notes |
|---|---|---|---|
| Claude 1 | 2023 | Discontinued | Early API assistant; superseded by Claude 2. |
| Claude 2 | Jul 2023 | Discontinued | 100K context class; first widely public chatbot. |
| Claude 3 Opus / Sonnet / Haiku | Mar 2024 | Superseded | First three-tier Claude 3 family; multimodal. |
| Claude 3.5 Sonnet / Haiku | 2024 | GA / legacy | Artifacts-era chat; strong coding; some SKUs retired per provider schedule. |
| Claude 3.5 Opus (where offered) | 2024–2025 | GA / legacy | Regional / product rollouts vary; check API IDs. |
| Claude 4 Opus / Sonnet | May 2025 | GA | Gen-4 numbering; agentic APIs and tool ecosystem push. |
| Claude Opus 4.1 | Aug 2025 | GA | Incremental Opus track; safety / conversation features in release notes. |
| Claude 4.5 Opus / Sonnet / Haiku | Sep–Nov 2025 | GA | 4.5 wave; Haiku 4.5 for speed tier. |
| Claude 4.6 Opus / Sonnet | Feb 2026 | GA | Current frontier; extended / adaptive thinking where listed above. |
Release chronology (major SKUs)
Reverse-chronological museum index — dates align with public announcements and doc rollouts. Anthropic model docs are the source of truth for IDs and deprecation.
| Model | Release | Docs |
|---|---|---|
| Claude Sonnet 4.6 | February 2026 | Models |
| Claude Opus 4.6 | February 2026 | Models |
| Claude Opus 4.5 | November 2025 | Models |
| Claude Haiku 4.5 | October 2025 | Models |
| Claude Sonnet 4.5 | September 2025 | Models |
| Claude Opus 4.1 | August 2025 | Models |
| Claude Sonnet 4 & Opus 4 | May 2025 | Models |
| Claude Sonnet 3.7 | February 2025 | Models |
| Claude Haiku 3.5 & Sonnet 3.5 (refresh) | October 2024 | Models |
| Claude Sonnet 3.5 | June 2024 | Models |
| Claude 3 family | March 2024 | Models |
| Claude 2 | July 2023 | Models |
Technical notes
| Topic | Public detail |
|---|---|
| Context windows | 200K-class tokens remain common; frontier 4.x SKUs may advertise 1M-class contexts on select platforms — always read the card for the ID you deploy. |
| Extended vs adaptive thinking | Extended thinking exposes longer reasoning chains; adaptive thinking adjusts depth dynamically — Haiku 4.5 may omit adaptive per docs. |
| Vision | Claude 3+ multimodal — image understanding in API and consumer apps. |
| Computer use | Agentic / desktop control features (product-specific) in 2024–2025; paired with MCP-style tool ecosystems in industry coverage. |
| MCP | Model Context Protocol (open connectors for tools and data) is widely used with Claude in IDEs; see modelcontextprotocol.io. |
Selected timeline (2024–2026)
- Mar 2024: Claude 3 Opus / Sonnet / Haiku — multimodal benchmark story.
- Mid–late 2024: Claude 3.5 Sonnet (June) and Haiku/Sonnet refresh (Oct); Artifacts and computer-use betas in product.
- Feb 2025: Claude Sonnet 3.7 — long reasoning & IDE narratives.
- May 2025: Claude 4 Sonnet & Opus — gen-4 numbering and tool ecosystem.
- Aug 2025 → Feb 2026: Opus 4.1 → 4.5 wave (including Haiku 4.5) → Opus / Sonnet 4.6.
- Ongoing: Thinking and extended-reasoning modes in Cursor-class IDEs; pricing and IDs tracked in docs.
References
Anthropic · Claude (language model) · Claude product · Models overview (docs) · Pricing
Museum compilation — not an official Anthropic model sheet. IDs, prices, and cutoffs change; confirm in Anthropic’s documentation before production use.