Claude Opus 4.6 Review: Benchmarks & Pricing (2026)
By Loïc Jané · Founder, Fleece AI
Claude Opus 4.6: Anthropic's February 2026 Flagship, Reviewed
At a Glance: Claude Opus 4.6 is an Anthropic model released February 5, 2026, with 128K output tokens, a 1M token context window, and, at launch, the top score on Terminal-Bench 2.0 for agentic coding. Anthropic now lists it as a legacy model and recommends Claude Opus 5.5, which costs less ($4/$20 per million tokens against $5/$25). On Fleece AI, Opus 4.6 is no longer selectable for new agents: agents already pinned to it keep running, and the Business plan offers Claude Opus 5.5, 4.8 and 4.7 instead. Reviewed February 20, 2026; updated September 29, 2026.
Update (September 2026): Claude Opus 4.6 is no longer Anthropic's most capable model. Anthropic has since released Claude Opus 4.7, Opus 4.8, Opus 5 and, on September 22, 2026, Claude Opus 5.5. Anthropic's documentation marks Opus 4.6 as "Active (legacy)", with retirement not sooner than February 5, 2027. It stays on the API at $5/$25 per million tokens, which is now more than Opus 5.5 ($4/$20). Claude Fable 5.1 ($10/$50) is priced above both. On Fleece AI, Opus 4.6 no longer appears in the model picker; the Business plan offers Claude Opus 5.5, 4.8 and 4.7. The benchmark claims below describe the February 2026 launch; Terminal-Bench 2.0 has since been replaced by Terminal-Bench 4.0 and SWE-bench Verified was retired.
Claude Opus 4.6 was Anthropic's most capable model when it was released on February 5, 2026. On Fleece AI, it is kept only for agents already pinned to it. For new workflows that need deep reasoning, long outputs and multi-step agentic work, the Business plan offers Claude Opus 5.5, 4.8 and 4.7.
What Is Claude Opus 4.6?
Claude Opus 4.6 succeeded Opus 4.5 and focused on coding, agents and enterprise knowledge work. Its main features at launch:
- Agent teams in Claude Code (research preview): several agents split a larger task and coordinate with each other
- 1M token context window: in beta at launch, now included at standard pricing
- 128K output tokens on the synchronous API (up to 300K on the Batch API with a beta header)
- Agentic coding: the highest score on Terminal-Bench 2.0 at launch, according to Anthropic
For a full breakdown, see our Best AI Models for Workflow Automation 2026 comparison.
Key Capabilities
Launch Benchmarks
In its February 2026 announcement, Anthropic reported that Claude Opus 4.6:
- achieved the highest score on Terminal-Bench 2.0, an agentic coding benchmark
- led all frontier models on Humanity's Last Exam, a multidisciplinary reasoning test
- scored about 144 Elo points above GPT-5.2 and 190 above Opus 4.5 on GDPval-AA, a test of economically valuable knowledge work
- performed better than any other model on BrowseComp, which measures finding hard-to-locate information online
- reached 81.42% on SWE-bench Verified with a prompt modification (its headline score was averaged over 25 trials)
Newer models have since overtaken it, and several of these benchmarks have been replaced. For what each benchmark measures today, see our AI Agent Benchmarks 2026 Explained guide.
128K Output Tokens
With 128K output tokens, Claude Opus 4.6 can produce long outputs in one pass:
- Comprehensive reports of dozens of pages
- Detailed code implementations
- Multi-section analyses with full supporting data
- Extended document drafts
Current Claude models, including Opus 5.5, have the same 128K limit.
Deep Document Understanding
The 1M token context window, now at standard pricing, suits Opus 4.6 to:
- Contract analysis and comparison
- Research paper summarization
- Regulatory compliance reviews
- Multi-document synthesis
Note that Opus 4.6 uses Anthropic's older tokenizer: 1M tokens holds about 750,000 words, against about 555,000 words on Opus 4.7 and later models.
Agent Teams (New in 4.6)
Claude Opus 4.6 introduced agent teams in Claude Code: several agents coordinate on one complex task and run independent parts in parallel. Fleece AI handles multi-step orchestration at the platform level, through agent hierarchies and delegation, whichever model an agent uses.
Claude Opus 4.6 on Fleece AI Today
Claude Opus 4.6 costs $5/$25 per million tokens (input/output) on Anthropic's API, more than its successor Claude Opus 5.5 ($4/$20). Fleece AI offered it on the Business plan, but it is no longer selectable for new agents on any plan; agents already pinned to it keep running. Claude models on Fleece AI are now Opus 5.5, 4.8 and 4.7 on Business, and Fable 5 on Enterprise.
| Plan | Available Models |
|---|---|
| Starter (€49/mo) | GLM 5.3 Flash (default), GPT-5.2, Mistral Medium 3.5, GPT-5.6 Luna, Qwen 3.8 Max, Kimi K3, DeepSeek V4.1 Flash |
| Pro (€99/mo) | All Starter models + GPT-5.6 Terra |
| Business (€199/mo) | All Starter and Pro models + GPT-5.6 Sol, Claude Opus 5.5, 4.8 and 4.7 |
| Enterprise (custom) | All Business models + Claude Fable 5 |
Best Use Cases on Fleece AI
These workflows suited Opus 4.6 and run on Claude Opus 5.5 today.
Deep Research and Analysis
"Every week, pull all new academic papers from arXiv mentioning 'AI safety', summarize each paper's methodology and findings, and compile a research digest with cross-paper insights. Send as a formatted email to the research team."
Deep reasoning and long outputs make Opus models a good fit for research synthesis.
Complex Document Processing
"When a new contract is uploaded to Google Drive, analyze it for key terms, obligations, renewal clauses, and liability provisions. Generate a structured summary with risk flags and post to Slack #legal."
Nuanced document understanding is one of Claude's strongest capabilities.
Code-Heavy Automation
"Every day, analyze our GitHub repository for open pull requests older than 3 days. For each, review the code changes, identify potential issues, and post a summary review comment on the PR."
Strong coding performance translates to reliable code analysis workflows.
Multi-Source Reporting
"Every month, compile data from Salesforce (pipeline), Stripe (revenue), HubSpot (leads), and Google Analytics (traffic). Generate a 10-page investor-ready monthly report with charts descriptions and commentary."
The 128K output limit lets the model write a full report in a single pass.
Claude Opus 4.6 vs Claude Opus 5.5
| Feature | Claude Opus 4.6 | Claude Opus 5.5 |
|---|---|---|
| Released | February 5, 2026 | September 22, 2026 |
| Status at Anthropic | Legacy (retirement not before Feb 5, 2027) | Current (retirement not before Sep 22, 2027) |
| Context window | 1M | 1M |
| Max output | 128K (300K on Batch API, beta) | 128K (300K on Batch API, beta) |
| API price (input/output) | $5/$25 per M | $4/$20 per M |
| Thinking | Adaptive (extended deprecated) | Adaptive (always on) |
| Default effort | high | medium |
| Reliable knowledge cutoff | May 2025 | June 2026 |
| Fleece AI | Pinned agents only | Business plan |
Source: Anthropic's Claude Opus 4.6 and models overview pages and the pricing page, retrieved September 29, 2026.
The two models were benchmarked on different test versions, so their scores do not compare directly. Opus 4.6 launched on Terminal-Bench 2.0 and SWE-bench Verified. For Opus 5.5, Anthropic reports 66.4% on Terminal-Bench 4.0 and 81.8% partial credit on OSWorld 2.1 (Anthropic, September 2026).
Run Claude Opus 5.5 instead — Upgrade to Fleece AI Business for Claude Opus 5.5, 4.8 and 4.7 on your deepest analysis workflows.
How to Use Claude Opus on Fleece AI
- Subscribe to the Business plan at fleeceai.app/subscribe
- Open any chat or flow
- Click the model selector
- Select Claude Opus 5.5 (or 4.8 or 4.7) and describe your workflow
Claude Opus 4.6 no longer appears in the model selector; agents already pinned to it keep running on it. Starter and Pro plans do not include Claude models.
Frequently Asked Questions
How does Claude Opus 4.6 compare to Claude Opus 5.5?
Both have a 1M token context window and 128K output tokens. Opus 5.5 is newer (September 22, 2026), has a more recent knowledge cutoff (June 2026 against May 2025) and costs less: $4/$20 per million tokens against $5/$25. Anthropic recommends migrating from Opus 4.6 to Opus 5.5.
Is Claude Opus 4.6 available on Fleece AI?
Not for new agents. It was previously offered on the Business plan; agents already pinned to it keep running. For new agents, the Business plan (€199/month) offers Claude Opus 5.5, 4.8 and 4.7.
Is Claude Opus 4.6 being retired?
Anthropic lists Opus 4.6 as a legacy model that is still available, with retirement not sooner than February 5, 2027 on the Claude API. Amazon Bedrock and Google Cloud set their own dates.
Can Claude Opus 4.6 generate full reports in one pass?
Yes. With 128K output tokens, Claude Opus 4.6 can write long reports in a single execution without splitting. Claude Opus 5.5 has the same limit.
Related Articles
- Claude Opus 4.7 Review — the model that followed Opus 4.6, still on Fleece AI Business
- Gemini 3.1 Pro vs Claude Opus 4.6 — head-to-head comparison
- Claude vs Fleece AI: Assistant vs Agent — Claude standalone vs Claude powering Fleece AI
- GPT-5.2 on Fleece AI — available on every plan
- Best AI Models for Workflow Automation 2026 — full comparison
Get Claude Opus 5.5 on Fleece AI — the Business plan includes Claude Opus 5.5, 4.8 and 4.7.