Overview
Claude 2.1 is a large language model released by Anthropic on November 21, 2023, as an incremental update to Claude 2.0. Its headline feature was a 200,000-token context window — roughly 150,000 words or 500+ pages — which let Claude 2.1 ingest long technical manuals, financial filings, codebases, or entire books in a single prompt. At launch this was the largest context window offered by a major commercial LLM. Claude 2.1 was available through the Anthropic API and Console and powered the claude.ai chat product, with the full 200K window reserved for Claude Pro subscribers.
Beyond the longer context, Claude 2.1 focused on trustworthiness. Anthropic reported a 2x decrease in false statements compared with Claude 2.0, a 30% reduction in incorrect answers on document-grounded questions, and a 3-4x lower rate of mistakenly concluding that a document supports a given claim. The release also introduced system prompts — custom instructions that set Claude's role, persona, or output format — and a beta tool-use capability that let developers wire Claude up to their own functions, APIs, calculators, web search, and private knowledge bases. Anthropic's own model card frames Claude 2.1 as offering the same underlying performance as Claude 2 plus these honesty and reliability gains.
Because Claude 2.1 shares the Claude 2 base model, its capability benchmarks come from the Claude v2 model card: 71.2% on Codex HumanEval (Python), 88.0% on GSM8K math, 78.5% on MMLU, and 91.0% on ARC-Challenge, with the model also reaching the ~95th percentile on the GRE Verbal section and scoring 76.5% on the Multistate Bar Examination. The model was text-only (no vision), had a knowledge cutoff in early 2023, and was priced at $8 per million input tokens and $24 per million output tokens. Claude 2.1 is now retired: Anthropic deprecated the claude-2.1 API model on January 21, 2025 and shut it down on July 21, 2025, pointing users to the Claude 3 family and later models such as Claude Opus 4.8.
| Released | 2023-11-21 |
|---|---|
| License | Proprietary |
| Weights | API only |
| Parameters | Undisclosed |
| Context | 200K tokens |
| Max output | ~4,000 tokens (Claude 2 was trained to generate coherent documents up to 4,000 tokens) |
| Architecture | Proprietary autoregressive transformer trained with unsupervised learning, RLHF, and Constitutional AI; shares the Claude 2 base, with Claude 2.1 adding reliability/honesty improvements over the full 200K context. |
| Knowledge cutoff | Early 2023 (training data cuts off in early 2023, per the Claude v2 model card) |
| Modalities | Text |
| Status | Retired — the claude-2.1 API model was deprecated on January 21, 2025 and retired on July 21, 2025 on Anthropic-operated platforms. Requests to it now fail; Anthropic's recommended replacement is Claude Opus 4.8. |
Benchmarks
- Codex HumanEval (Python, 0-shot, Claude 2 base)71.2%
- GSM8K (0-shot CoT, Claude 2 base)88%
- MMLU (5-shot CoT, Claude 2 base)78.5%
- ARC-Challenge (5-shot, Claude 2 base)91%
- Multistate Bar Exam / MBE (5-shot, Claude 2 base)76.5%
Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.
Pricing
| Input | $8.00 per million tokens |
|---|---|
| Output | $24.00 per million tokens |
Anthropic's published November 2023 pay-as-you-go pricing for Claude 2.1 (and Claude 2.0); the model is now retired and no longer purchasable.
Strengths
- Industry-leading 200K-token context window at launch — roughly 150,000 words or 500+ pages in a single prompt
- 2x fewer false statements than Claude 2.0 and a 30% reduction in incorrect answers on document-grounded questions
- 3-4x lower rate of wrongly concluding a document supports a claim, improving long-document faithfulness
- Introduced system prompts for setting role, persona, and output format
- Added a beta tool-use API for connecting Claude to developer functions, APIs, search, and knowledge bases
- Strong exam and reasoning results inherited from the Claude 2 base (76.5% MBE bar exam, ~95th-percentile GRE Verbal)
Best for
- Long-document analysis: summarizing and querying lengthy manuals, contracts, and financial filings
- Comparing or cross-referencing multiple documents or whole books in one prompt
- Codebase question-answering over large source files within the 200K window
- Retrieval-augmented and agentic workflows via the beta tool-use API
- Drafting, editing, and outlining with system-prompt-defined personas and formats
- Knowledge and reasoning Q&A where reduced hallucination matters
How to access
| Provider | Model ID |
|---|---|
| Anthropic ↗ | claude-2.1 |
Claude (legacy 1–2.x) — every version
The full lineage of the Claude (legacy 1–2.x) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.
| Version | Released | Context | License |
|---|---|---|---|
| Claude 2.1current | 2023-11-21 | — | Proprietary |
| Claude 2 | 2023-07-11 | — | Proprietary |
| Claude Instant | 2023-03-14 | — | Proprietary |
| Claude 1 | 2023-03-14 | — | Proprietary |
FAQ
When was Claude 2.1 released and what was new?
Anthropic released Claude 2.1 on November 21, 2023. It added a 200K-token context window, system prompts, and a beta tool-use API, and reduced false statements by about 2x compared with Claude 2.0.
How big is Claude 2.1's context window?
200,000 tokens — roughly 150,000 words or over 500 pages. At launch this was the largest context window of any major commercial LLM, and the full 200K window was available to Claude Pro and API users.
How much did Claude 2.1 cost?
Per Anthropic's published November 2023 pricing, Claude 2.1 cost $8.00 per million input tokens and $24.00 per million output tokens. The model is now retired and no longer available to purchase.
Is Claude 2.1 still available?
No. The claude-2.1 API model was deprecated on January 21, 2025 and retired on July 21, 2025 on Anthropic-operated platforms. Requests to it now fail; Anthropic recommends migrating to Claude Opus 4.8 or another current model.