Claude Opus 4.6: 1M Context Window Is Now the Default — 5x More Room, Same Price
Anthropic's Claude Opus 4.6 ships with a 1M-token context window by default — up from 200K in Opus 4.5. Same $5/$25 per MTok pricing. Here's what changes for developers.
Claude Opus 4.6 now ships with a 1 million token context window by default. No beta headers. No special flags. Just 1M tokens, ready to use.
For comparison, Opus 4.5 had a 200K-token context window. That’s a 5x increase at the same $5 / $25 per MTok pricing.
The Numbers
| Opus 4.5 | Opus 4.6 | Change | |
|---|---|---|---|
| Context window | 200K tokens | 1M tokens | 5x |
| Max output | 64K tokens | 128K tokens | 2x |
| Input pricing | $5 / MTok | $5 / MTok | Same |
| Output pricing | $25 / MTok | $25 / MTok | Same |
| Extended thinking | Yes | Yes | — |
| Adaptive thinking | No | Yes | New |
Sonnet 4.6 also defaults to 1M tokens (previously required a beta header on Sonnet 4.5). Haiku 4.5 stays at 200K.
What 1M Tokens Actually Means
One million tokens is roughly:
- ~750,000 words — the entire Harry Potter series is ~1.08M words
- ~3.4 million Unicode characters
- An entire medium-sized SaaS codebase loaded at once
- 100+ technical documents in a single conversation
For Claude Code users, this is the most impactful change. Your entire project directory — source files, tests, configs, documentation — can fit in a single context window without splitting or summarizing.
What Changes for Developers
1. No More Context Management Gymnastics
With 200K tokens, developers had to carefully manage what went into the context:
- Splitting large codebases into chunks
- Summarizing documentation before injecting it
- Carefully selecting which files to include
- Dealing with “context full” interruptions mid-task
With 1M tokens, most projects simply fit. The cognitive overhead of context management drops to near zero for typical workloads.
2. Agents Run Longer
AI agents that loop through tool calls accumulate context with every step. At 200K, agents would hit the ceiling mid-task and need to compact — losing information in the process.
At 1M tokens, agents can sustain ~5x more iterations before hitting limits. For Claude Code’s autonomous workflows, this means:
- Multi-file refactors complete without interruption
- Complex debugging sessions maintain full history
- Long-running automation loops stay coherent
3. Max Output Doubles to 128K
Opus 4.6 also doubles max output from 64K to 128K tokens. This means:
- Longer code generation in a single response
- Complete file rewrites without truncation
- More detailed analysis and explanations
4. Adaptive Thinking (New)
Opus 4.6 introduces adaptive thinking — the model dynamically adjusts its reasoning depth based on task complexity. Simple questions get fast answers; complex problems get deeper analysis. This was previously unavailable on Opus models.
Pricing Context
The pricing stays flat at $5 input / $25 output per million tokens. But the effective value increases dramatically:
| Metric | Opus 4.5 | Opus 4.6 |
|---|---|---|
| Cost per token of context | Same | Same |
| Usable context | 200K | 1M |
| Max output per request | 64K | 128K |
| Effective capacity | 1x | 5x |
For batch API users, the standard 50% discount still applies: $2.50 input / $12.50 output per MTok.
Long Context Pricing Note
For older models (Sonnet 4.5, Sonnet 4), using 1M context required a beta header and incurred long context pricing for requests exceeding 200K tokens. With Opus 4.6 and Sonnet 4.6, 1M is the native context window — standard pricing applies throughout.
Full Model Lineup (March 2026)
| Model | Context | Max Output | Input $/MTok | Output $/MTok |
|---|---|---|---|---|
| Opus 4.6 | 1M | 128K | $5 | $25 |
| Sonnet 4.6 | 1M | 64K | $3 | $15 |
| Haiku 4.5 | 200K | 64K | $1 | $5 |
| Opus 4.5 (legacy) | 200K | 64K | $5 | $25 |
| Sonnet 4.5 (legacy) | 200K (1M w/ beta) | 64K | $3 | $15 |
Context Quality: Not Just Bigger, But Better
Previous long-context models suffered from “context rot” — performance degraded as the context filled up. Opus 4.6 addresses this directly.
On the MRCR v2 benchmark (a needle-in-haystack retrieval test at 1M tokens):
- Opus 4.6: 76% accuracy
- Sonnet 4.5: 18.5% accuracy (same benchmark)
That’s a 4x improvement in long-context retrieval. The model doesn’t just have more room — it actually uses that room effectively.
Additional Improvements
- Media limit increased: Up to 600 images or PDF pages per request (vs. 100 with 200K context)
- Context compaction (beta): Server-side automatic summarization enables effectively infinite conversations
- Fine-grained tool streaming: Stream tool parameters without JSON buffering, reducing latency in multi-tool workflows
Who Benefits Most
- Claude Code users — Entire project directories fit in context. No more splitting tasks.
- Agent builders — Longer autonomous runs without context compaction.
- Document processing — Ingest full codebases, legal documents, or research papers in one pass.
- RAG systems — Larger retrieval windows mean fewer chunks and better coherence.
Getting Started
Claude Code Availability
Opus 4.6 with 1M context is now the default model for Claude Code users on:
- Max plan — default
- Team plan — default
- Enterprise plan — default
- Pro plan — opt in with
/extra-usage - Sonnet users — opt in with
/extra-usage
For API users, the model ID is claude-opus-4-6. The 1M context is automatic — no configuration needed.
# API call with Opus 4.6 — 1M context is automatic
curl https://api.anthropic.com/v1/messages \
-H "x-api-key: $ANTHROPIC_API_KEY" \
-H "content-type: application/json" \
-H "anthropic-version: 2023-06-01" \
-d '{
"model": "claude-opus-4-6",
"max_tokens": 4096,
"messages": [{"role": "user", "content": "Hello"}]
}'No beta headers. No extended context flags. Just use it.
For complete pricing details, see the official pricing page. For model comparison, see Models overview.