AI tools·Chat models·Global
Claude
Known for writing quality, long context, and coding agents. Claude Code, Artifacts, and computer use are the differentiators.
Compiled through 2026-09
- Vendor
- Anthropic
- Started
- 2021
- China availability
- Overseas accessRequires overseas network access
- Pricing tier
- Free tier
- Current flagship
- Claude 3.7 Sonnet / Fable 5.1 / Opus 5
- Website
- claude.ai
Evaluation, Pros & Cons
Editorial score based on public benchmarks and community feedback — not a third-party benchmark result
Key Advantages
- Claude 3.7 Sonnet hybrid reasoning leads coding and logical rigor globally
- Superb Artifacts UX with exceptional long-context adherence and low hallucination
Limitations & Caveats
- Free tier is restrictive and regional access requires proxy
Overview
Anthropic started from Constitutional AI. Claude is strong on control, long documents, and engineering. Surfaces include claude.ai, the API (Bedrock / Google Cloud / Azure), Claude Code, and computer use. The Claude 4 line in 2025 pushed coding agents; July 2026 added Opus 5, Sonnet 5, and Fable 5, then Fable 5.1 in September with a 1M-token default context.
Direction
Public focus is long-horizon agents (Fable / Mythos), million-token context, computer use, and Claude Code. Sonnet stays the workhorse; Opus handles hard reasoning. Cloud-vendor distribution continues. Mythos remains limited (e.g. Project Glasswing).
Version history
Sep 2026
LatestClaude Fable 5.1
1M context by default; adaptive thinking.
View details
New capabilities
- 1M-token context by default
- Adaptive thinking: less on easy asks, more on hard ones
- Aimed at huge repos and long refactors
What got stronger
- No beta form to stuff a whole service into context
- Fewer mid-run “forgot the goal” failures
Version notes
Fable 5.1 made 1M context the default and added adaptive thinking. The feel is: drop a monorepo and a fuzzy refactor, and it decides how long to think. Quotas and official docs still win.
Jul 2026
Opus 5 / Sonnet 5 / Fable 5
Launched Jul 24; Fable for long-running agents.
View details
New capabilities
- Opus 5 / Sonnet 5 as the new quality/value split
- Fable 5 for long-horizon coding agents
- Shipped together on 24 July
What got stronger
- Fable targets jobs that run tens of minutes to hours
- Sonnet 5 stays daily; Opus 5 takes the hardest reasoning
Version notes
On 24 July 2026 Anthropic shipped the 5-generation trio. Fable is the new name that matters: not faster chat, but splitting an engineering goal, running it, and self-correcting. Claude Code is a primary surface.
May 2026
Opus 4.6–4.8 / Sonnet 4.6
Coding, vision, 1M context (beta).
View details
New capabilities
- 1M-token context entered beta
- Better vision and coding together
What got stronger
- Huge repos and logs could be dropped in more whole
Version notes
Opus 4.6–4.8 and Sonnet 4.6 were the bridge before the 5 line: 1M context, vision, and coding on the models customers already had. Useful if you were already on the Claude API and did not want to wait for Fable.
Nov 2025
Claude Opus 4.5 / Haiku 4.5
Late 4.x upgrades.
View details
New capabilities
- Opus 4.5 closed the 4.x flagship line
- Haiku 4.5 brought the fast model up to useful work
What got stronger
- Flagship reasoning up a notch; the fast SKU could do light agents
Version notes
4.5 was the late 4.x bump: Opus for a last intelligence tick, Haiku so cheap SKUs could do light agents. 2026’s 4.6–4.8 and the 5 line then made million-token context the default story.
Sep 2025
Claude Sonnet 4.5
Real-world agents and coding.
View details
New capabilities
- Real-world agents: browser, files, long jobs
- Another jump in coding benches and stability
What got stronger
- Better at “here’s a repo, go” at the same price tier
- Fewer dropped constraints in long chats
Version notes
Sonnet 4.5 was positioned as the real-world agent workhorse: not always #1 on contest benches, but steadier at editing code, clicking UIs, and keeping project rules. Many teams made it the default coding model.
May 2025
Claude 4 / Opus 4
New flagship for agents and code.
View details
New capabilities
- Claude 4 / Opus 4 aimed at agents and repo-scale code
- Longer autonomous runs and tighter tool loops
- Paired with the Claude Code terminal agent
What got stronger
- Repo edits stalled less often than on 3.5
- Plan-run-fix loops were more complete
Version notes
Claude 4 shifted the story from great writing to working like an engineer. Opus 4 for hard reasoning, Sonnet 4 for daily agents. Claude Code became a first-class terminal citizen at the same time.
Jun 2024
Claude 3.5 Sonnet
Jump in coding and Artifacts.
View details
New capabilities
- Artifacts: previewable code/doc canvas
- Jump in coding and small frontend apps
- Groundwork for computer use
What got stronger
- Better React and spreadsheet work than daily 3 Opus
- Artifacts made generate-and-see a Claude.ai feature
Version notes
3.5 Sonnet was a rare mid-tier that beat the old flagship on a lot of coding and writing, cheaper too. Artifacts turned output into a clickable page instead of a chat bubble.
Mar 2024
Claude 3 family
Opus / Sonnet / Haiku tiers.
View details
New capabilities
- Opus / Sonnet / Haiku price tiers
- 200K-class long context as a headline
- Tool / function calling
What got stronger
- Haiku for cheap volume; Opus for hard problems
- Long PDFs, contracts, and repos started to fit
Version notes
The Claude 3 family made “one brand, three SKUs” the industry template. Opus for intelligence, Sonnet for daily work, Haiku for latency and cost. Long context started beating GPT-4’s window on documents.
Mar 2023
Claude 1
First chat and API generation.
View details
New capabilities
- Chat product and API launched together
- Safer, more restrained outputs
What got stronger
- Writing quality started to differentiate
- Safer tone made enterprise drafts more acceptable
Version notes
Claude 1 was Constitutional AI as a product. It was not yet Opus-class, but the calmer writing voice was already the brand. Chat and API shipped together so developers could try it.