When to open: You want to look up "When was Claude X.Y released / What's different from the previous generation / Should I upgrade / Is Max Plan worth it / Which SKU fits my work". This wiki is the official-style timeline of 12 events in Anthropic's 2024–2026 product releases, plus the key selling points of each generation.
One-sentence core:Anthropic's product cadence is "3 SKUs (Opus/Sonnet/Haiku) × half-year generations (4.x) + intermediate decimal-point iterations (.5/.6/.7)" — unlike OpenAI's confusing naming of "4 / 4.5 / 4o / o1 / o3", Anthropic uses semantically clear family + version numbers, with the explicit intent that "users don't have to guess which is strongest every time a release happens".
Boundary with wangkai / newtype model evaluations: This wiki takes the official release perspective(how Anthropic presents itself);for hands-on evaluations + cross-model comparisons see newtype Model Evaluations, Fine-tuning, and Hardware Evolution / Wang Kai's AI Industry Judgments and Observations on Big Tech Competitive Landscape.
0. Overview of 9 posts + 3 landing pages
| Time | Event | Type |
|---|---|---|
| 2025-04 | Max Plan launch (20x usage) | Commercial |
| 2025-09 | Claude Sonnet 4.5 release | Model |
| 2025-10 | Claude Haiku 4.5 release | Model |
| 2026-02 | Claude is a space to think (positioning statement) | Positioning |
| 2026-02 | Claude Sonnet 4.6 release (including 1M context beta) | Model |
| 2026-04 | Claude Opus 4.7 release | Model |
| 2026-04 | New Claude Design product (Anthropic Labs) | Product |
| 2026-05 | Higher limits + SpaceX compute partnership | Commercial + Compute |
| (ongoing) | claude.com/opus / sonnet / haiku product landing pages | Product page |
1. Model generation timeline (simplified)
| Generation | Release | Main focus |
|---|---|---|
| Claude 3 Opus | 2024-03 | Strong reasoning |
| Claude 3.5 Sonnet | 2024-06 | Coding momentum + cost-effectiveness |
| Claude 3.5 Haiku | 2024-11 | Small model |
| Claude 3.7 Sonnet | 2025-02 | (Economic Index report period) |
| Claude 4 Opus | 2025-05 | Internal activation of ASL3 (Complete Anthology of Anthropic AI Safety / Alignment / Policy Research (Constitutional AI + RSP + Welfare + Policy) §2.2) |
| Claude 4.6 Opus / Sonnet | 2025-09~2026-02 | (mid-generation) |
| Claude Sonnet 4.5 | 2025-09-29 | "best coding model in the world" / "best for building complex agents" / "best at using computers" |
| Claude Haiku 4.5 | 2025-10-15 | small / affordable / coding capability close to the top-tier Sonnet 4 from 5 months prior |
| Claude Sonnet 4.6 | 2026-02-17 | Comprehensive upgrade + 1M token context window beta |
| Claude Opus 4.7 | 2026-04-16 | Advanced software engineering / hard tasks you can "hand off" |
| (next) | 2026-Q3+ | 5.0 series / extended thinking as default, etc. |
Pattern: Each major Claude X generation lasts roughly 9–12 months, during which.5 /.6 /.7 minor versions arrive once every 2–4 months. Opus usually lags Sonnet by 1–2 months (Opus is the strongest but needs time to polish).
2. Key points of each release (my condensation)
2.1 Claude Sonnet 4.5(2025-09-29)
Anthropic's own words:
- "best coding model in the world"
- "best for building complex agents"
- "best at using computers" (strongest computer use)
- Reasoning + math significantly improved
Significance:Sonnet replaces Opus as the default workhorse for daily work — Opus is reserved for "extremely hard tasks", while Sonnet offers better cost-effectiveness.
2.2 Claude Haiku 4.5(2025-10-15)
Anthropic's own words(simplified):
- Coding capability close to Sonnet 4 (the top tier from 5 months prior)
- Price + speed advantages on both ends
- Suitable for large volumes of light tasks / service backends / client-side edge scenarios
Significance:The top-tier model from 5 months ago becomes the small model 5 months later — this is a typical phenomenon in the LLM industry, where frontier capability rapidly cascades down to commodity.
2.3 Claude is a space to think (2026-02-04 positioning statement)
Key quote(original text):
"There are many good places for advertising. A conversation with Claude is not one of them."
Translation: There are many places where ads make sense. A conversation with Claude is not one of them.
Anthropic's stance:Claude will never run ads, which forms an explicit contrast with ChatGPT's ad integration / Google Search's business model.
Significance:
- On the business model side it relies on API + subscriptions + enterprise plans(not ads)
- to maintain information credibility(ads distort recommendations)
- Cross-checks with Anthropic Interpretability + Claude Behavior Research (Natural Language Autoencoders + User Usage Insights) §2 "personal guidance" research — 6% of conversations are seeking life advice,and ads would be very dangerous in such scenarios
2.4 Claude Sonnet 4.6(2026-02-17)
Anthropic's own words:
- "Most capable Sonnet model yet"
- Comprehensive upgrade: coding / computer use / long-context reasoning / agent planning / knowledge work / design
- 1M token context window(beta) —— 2x the previous, large codebases can be loaded in one go
Significance:
- Long-running agent scenarios benefit the most(larger context reduces the need for compaction, see Anthropic Engineering Team First-Party Perspective: Agent Engineering / Skills / MCP / Claude Code (2025–2026 Complete Collection) §3.4 managed agents)
- Marks Anthropic's entry into the "million-token context default" era
2.5 Claude Opus 4.7(2026-04-16)
Anthropic's own words(simplified):
- "Notable improvement on Opus 4.6"
- Main advantages:advanced software engineering(advanced software engineering),hardest tasks show particular improvement
- User feedback: the hardest coding work that previously required "close supervision" can now be "hand off"
Significance:Opus is the model for "I close my eyes and hand it off", while Sonnet is the model for "I watch while I work". The positioning difference between the two is becoming increasingly clear.
2.6 Max Plan(2025-04-09)
Anthropic commercial:
- 20x the previous usage limits
- Priority access to new features
- Price point / positioning sits between Pro and Team
Significance:Heavy individual users don't need to upgrade to enterprise; there is a dedicated tier. It addresses the real pain point that agentic workflows such as Claude Code "burn through a lot of tokens".
2.7 Higher limits + SpaceX compute partnership (2026-05-06)
Event: Anthropic signed a compute partnership with SpaceX, significantly expanding compute capacity → user limits raised.
Significance:
- Compute is the hard bottleneck for frontier AI, see Scaling Law / AGI Timelines / Software 3.0 (cross-author topic synthesis) / Wang Kai RSI AI Investment Strategy and the "Good Asset, Good Price, Good Volatility" Framework
- Partnering with SpaceX(rather than relying solely on AWS / Google Cloud) = diversification + securing compute supply
- The previous similar move was Microsoft + OpenAI's deep binding — Anthropic is not betting on a single cloud
2.8 Claude Design (2026-04-17, new Anthropic Labs product)
New product: Claude Design = collaborating with Claude on visual work(design, prototypes, UI, etc.).
Placement: Under Anthropic Labs (an independent product line, parallel to Claude / Cowork).
Significance:Anthropic expands from pure LLM services to "AI-native tools" — moving in the same direction as OpenAI Sora / Microsoft Designer, but taking its own design route. For details see Claude Cowork / Dispatch / Design / Tutorials + Use Cases (Complete User Operation Layer) §X Using Claude Design for demos / UX tutorials.
3. 3 product family landing pages (simplified)
From the 3 product landing pages at anthropic.com/claude/{opus,sonnet,haiku}:
| Family | Positioning (official page) | When to use |
|---|---|---|
| Opus | "Our most powerful and intelligent model" | Extremely hard tasks, hands-off collaboration, complex reasoning |
| Sonnet | "Best performance for everyday tasks" | Daily workhorse, coding, agentic workflows |
| Haiku | "Our fastest and most affordable model" | Large volumes of light tasks, backend services, edge scenarios |
Tradeoffs among the 3 SKUs:Speed↑ Cost↓(Haiku)vs Capability↑ Cost↑(Opus). Sonnet is the most commonly used middle ground, the "king of cost-effectiveness".
4. Cross-perspective comparison: Anthropic vs OpenAI vs Google product cadence
| Dimension | Anthropic | OpenAI | |
|---|---|---|---|
| Naming clarity | High(Opus/Sonnet/Haiku × 4.5/4.6/4.7) | Low (GPT-4 / 4o / o1 / o3-mini etc., confusing) | Medium (Gemini 1.5 / 2.0 / Pro / Flash) |
| Generation cadence | 9–12 month major versions + 2–4 month minor versions | Irregular | Around 6 months |
| Business model | No ads(made explicit in 2026-02) | Some ads | Deeply tied to search ads |
| Compute supply | Multi-cloud + SpaceX (diversified) | Deeply bound to Microsoft | In-house (mainly TPU) |
| Model retention | deprecate ≠ delete(see Complete Anthology of Anthropic AI Safety / Alignment / Policy Research (Constitutional AI + RSP + Welfare + Policy) §4.2) | Old GPT models are often shut down directly | Old versions are mostly shut down |
For detailed evolution see AI Industry Judgments and Company Competition (cross-author topic synthesis) / newtype AI Business, Industry, and Investment Observations.
5. Related wikis / cross-reference map
6. Meta-information
Rewriting Strategy: 9 product news articles + 3 family landing pages reorganized by "timeline + key points per release + cross-perspective comparison".The main focus is not on the benchmark details of each model(those will become outdated; see the practical evaluations in newtype Model Evaluations, Fine-Tuning, and Hardware Evolution / Wang Kai's Observations on AI Industry Judgments and the Competitive Landscape of Large Companies), but on official positioning + business model + naming cadence(these are the stable product narratives).
Bilingual Strategy(translation_status: bilingual):
- Chapter titles use mixed Chinese and English
- Retain original Anthropic quotes +
*译:中文* - Model generation tables retain English version numbers
Sources and Related Materials
- https://www.anthropic.com/news/claude-opus-4-7
- https://www.anthropic.com/news/claude-sonnet-4-6
- https://www.anthropic.com/news/claude-sonnet-4-5
- https://www.anthropic.com/news/claude-haiku-4-5
- https://www.anthropic.com/news/max-plan
- https://www.anthropic.com/news/higher-limits-spacex
- https://www.anthropic.com/news/claude-is-a-space-to-think
- https://www.anthropic.com/news/claude-design-anthropic-labs
- https://www.anthropic.com/claude/opus
- https://www.anthropic.com/claude/sonnet
- https://www.anthropic.com/claude/haiku
- First-Party Perspective from the Anthropic Engineering Team: Agent Engineering / Skills / MCP / Claude Code (Complete Collection 2025–2026)
- AI Agent Evaluation Methodology (Anthropic Engineering Team demystifying-evals Deep Dive)
- AI Industry Judgments and Company Competition (Cross-Author Thematic Synthesis)
- newtype Model Evaluations, Fine-Tuning, and Hardware Evolution
- Wang Kai's Observations on AI Industry Judgments and the Competitive Landscape of Large Companies