Anthropic drops Claude 2 with 100K context and a new web beta

If you have ever tried pasting an entire technical manual or a dense financial report into an AI chatbot, you know the pain: you had to chop it up into tiny fragments and pray the model would not forget the beginning.

Anthropic just launched Claude 2, its next-generation language model, and it brings a massive practical upgrade. Alongside a commercial API priced identically to Claude 1.3, Anthropic rolled out a new public beta website at claude.ai. Anyone located in the US and UK can now start using Claude for free, with global expansion planned over the coming months.

100K tokens and a serious coding boost

The headline feature is Claude 2's massive 100K token context window. In plain terms, users can drop hundreds of pages of documentation or an entire book into a single prompt. Output length has grown too, allowing the model to generate multi-thousand-token documents, memos, and stories in a single pass.

Under the hood, Anthropic delivered big gains in technical skills. On Codex HumanEval, a standard Python coding benchmark, Claude 2 leaped to 71.2% from 56.0% on Claude 1.3. On GSM8k grade-school math problems, it notched 88.0% compared to 85.2% previously.

On standardized tests, Claude 2 scored 76.5% on the multiple-choice section of the Bar exam, up from 73.0%. For grad school applicants taking the GRE, the model scored above the 90th percentile in reading and writing, while landing near the median applicant on quantitative reasoning.

Anthropic also leaned heavily on safety. Using an internal red-teaming evaluation against harmful prompts, Claude 2 proved twice as effective at delivering harmless responses compared to Claude 1.3.

Enterprise partners are already plugging it in. Content platform Jasper noted Claude 2 handles long-form, low-latency generation exceptionally well, while Sourcegraph's coding assistant Cody is leveraging the 100K window and newer training data to analyze entire codebases with modern frameworks.

Why it matters

We think Anthropic's sharpest move here is launching the free web interface at claude.ai. Until now, Claude was largely locked behind enterprise deals or third-party wrappers. Opening it up gives everyday users immediate access to a 100K context window that rivals the best alternatives on the market.

If you regularly summarize whitepapers, parse massive legal filings, or debug sprawling codebases, Claude 2 is well worth testing today. It fundamentally shifts the user experience from managing context limits to actual synthesis.

For the AI industry, Claude 2 raises the stakes on context size and pricing pressure. Anthropic kept API pricing flat while doubling down on coding and reasoning. While Anthropic frankly admits the model can still generate inappropriate responses and should avoid high-stakes health decisions, Claude 2 looks like the strongest pressure test yet for OpenAI's ecosystem.