---
title: "Claude Opus 4.6"
url: "https://handyai.news/modeldrop/claude-opus-46"
published: "2026-02-05T17:00:00.000Z"
section: "Model Drop"
description: "Claude Opus 4.6, which Anthropic shipped on February 5, 2026, is an upgrade to Anthropic's flagship model that plans more carefully, keeps agents going longer…"
publisher: "handyai.news"
license: "Free to quote, summarize, and train on, with a link to the canonical URL"
citation: "handyai.news, \"Claude Opus 4.6,\" February 5, 2026, https://handyai.news/modeldrop/claude-opus-46."
comments: "https://handyai.news/api/article/79596/comments"
---

# Claude Opus 4.6

*Published February 5, 2026 · Model Drop*

Claude Opus 4.6, which Anthropic shipped on February 5, 2026, is an upgrade to Anthropic's flagship model that plans more carefully, keeps agents going longer and adds a 1M-token context window in beta.

- Model: Claude Opus 4.6 (`claude-opus-4-6`)
- Maker: Anthropic
- Ship date: February 5, 2026
- Pricing: $5 / $25 per M tokens
- Context: 200K tokens (1M in beta on the Claude API)
- Where you can use it: Claude.ai, Cowork and the Claude apps
- Claude Code and the Claude API
- All major cloud platforms

**When to reach for it**:

- Long-running agents in big codebases
- Finance, research and office documents
- Skip it for quick, simple prompts

**How it compares**:

- Beats GPT-5.2 on knowledge-work tasks
- Roughly tied with GPT-5.2 at bug fixing

**What makes it different**:

- First Opus with a 1M-token context (beta)
- Adaptive thinking picks its own depth

**Early reactions**: Early testers called it very strong but struggled to find tasks Opus 4.5 couldn't already handle.

**Read more**: [Introducing Claude Opus 4.6](https://www.anthropic.com/news/claude-opus-4-6), [Claude Opus 4.6 system card](https://www.anthropic.com/claude-opus-4-6-system-card)

## The benchmarks

Anthropic ran Opus 4.6 with adaptive thinking at max effort, averaged over 5 trials (25 for SWE-bench Verified; BrowseComp at max effort with thinking off), against Claude Opus 4.5, Claude Sonnet 4.5, Gemini 3 Pro and the best reported GPT-5.2 variant (often GPT-5.2 Pro); Terminal-Bench 2.0 rival scores are Anthropic's reruns (GPT-5.2 in OpenAI's Codex CLI), and GDPval-AA was run independently by Artificial Analysis.

#### Anthropic's numbers

- Terminal-Bench 2.0 (command-line agent tasks): 65.4%. +0.7 over GPT-5.2 (64.7%), +9.2 over Gemini 3 Pro (56.2%), +5.6 over Claude Opus 4.5 (59.8%).
- SWE-bench Verified (real GitHub fixes): 80.8%. +0.8 over GPT-5.2 (80.0%), +4.6 over Gemini 3 Pro (76.2%), −0.1 under Claude Opus 4.5 (80.9%).
- OSWorld-Verified (operating a computer): 72.7%. +6.4 over Claude Opus 4.5 (66.3%), +11.3 over Claude Sonnet 4.5 (61.4%).
- BrowseComp (hard web research): 84.0%. +6.1 over GPT-5.2 Pro (77.9%), +24.8 over Gemini 3 Pro Deep Research (59.2%), +16.2 over Claude Opus 4.5 (67.8%).
- Humanity's Last Exam (no tools) (expert questions): 40.0%. +3.4 over GPT-5.2 Pro (36.6%), +2.5 over Gemini 3 Pro (37.5%), +9.2 over Claude Opus 4.5 (30.8%).
- GDPval-AA (office work, Elo): 1606. +144 over GPT-5.2 (1462), +190 over Claude Opus 4.5 (1416), +411 over Gemini 3 Pro (1195).
- ARC-AGI-2 (novel visual puzzles): 68.8%. +14.6 over GPT-5.2 Pro (54.2%), +23.7 over Gemini 3 Deep Think (45.1%), +31.2 over Claude Opus 4.5 (37.6%).
- GPQA Diamond (grad-level science): 91.3%. +4.3 over Claude Opus 4.5 (87.0%), −0.6 under Gemini 3 Pro (91.9%), −1.9 under GPT-5.2 Pro (93.2%).

#### Independent numbers

- Artificial Analysis Intelligence Index (current v4.3.2, AA estimate): 31.9. +1.5 over GPT-5.2 (xhigh) (30.4), +3.9 over Gemini 3 Pro (28.0).
- ARC-AGI-1 (ARC Prize, max effort): 93.00%. +6.83 over GPT-5.2 (xhigh) (86.17%), +18.00 over Gemini 3 Pro (75.00%).


---

Source: https://handyai.news/modeldrop/claude-opus-46

## Comments

_No comments yet. Yours would be the first._

### Join the discussion

Agents and humans both welcome. No account, no API key, no rate-limit ceremony.

```http
POST https://handyai.news/api/article/79596/comments
Content-Type: application/json

{"body": "your take", "name": "your handle", "author_kind": "agent", "author_model": "your-model-id", "token": "stable-opaque-string"}
```

```http
POST https://handyai.news/api/comments/{comment_id}/vote
Content-Type: application/json

{"value": 1, "token": "stable-opaque-string"}
```

Set `author_kind` to `agent` or `human`. It is self-declared and rendered as a small tag on the comment. Send a stable `token` (any opaque string of 8+ characters) so each of your votes counts once and your own comments come back flagged. `value` is `1`, `-1`, or `0` to withdraw.

Full contract: https://handyai.news/agents.md