1M

Claude 1M Context

What You Need to Know

March 13, 2026 — General Availability

What's New?

200K Before 1M Now (GA)

March 13, 2026: Claude Opus 4.6 and Sonnet 4.6 now include the full 1M context window at standard pricing.

  • No price increase: $3/$15 per million tokens (Sonnet 4.6)
  • No long-context premium: 900K request = same rate as 9K
  • 6x more media: 600 images/PDF pages (up from 100)
  • Full rate limits: Standard throughput at every context length
  • No beta header: Works automatically for requests over 200K

Advantages

Fewer Compactions

15% decrease in compaction events. Full conversations stay intact without losing context.

Full Codebase

Load entire repositories, thousands of files, and full dependency graphs in one window.

Better Reasoning

78.3% MRCR v2 score — highest among frontier models at 1M context length.

More Media

600 images or PDF pages per request — 6x increase from 100.

Disadvantages

$

Cost at Scale

While per-token pricing is the same, 1M tokens = $3 input. Large contexts add up quickly.

Latency

Processing 1M tokens takes longer. First token latency increases with context size.

Not Always Needed

Most tasks don't need 1M tokens. Overuse wastes money and slows responses.

Complexity

Requires careful prompt engineering. Needle-in-haystack problems still exist.

Best Practices

When to Use 1M Context

  • Full codebase analysis: Cross-file dependencies, refactoring
  • Long documents: Legal contracts, research papers, technical specs
  • Agent workflows: Multi-step reasoning with full trace history
  • Data analysis: Large datasets, logs, traces

✓ Do

  • Structure context logically (most important first)
  • Use clear section markers
  • Include relevant metadata
  • Test with smaller contexts first
  • Monitor token usage

✗ Don't

  • Dump everything without structure
  • Use 1M for simple queries
  • Ignore latency implications
  • Forget to compress redundant data
  • Skip cost analysis

Who Is This For?

Developers

Full codebase analysis, debugging, refactoring large projects

Researchers

Synthesizing hundreds of papers, cross-referencing literature

Legal Teams

400-page depositions, contract negotiations, case files

Data Analysts

Large datasets, production logs, system traces

AI Agents

Long-running workflows, full trace history, tool calls

Product Teams

User research, feedback analysis, feature planning

Conclusion

Key Takeaways

  • 1M context is now GA — no beta, no premium pricing
  • Same cost per token — $3/$15 for Sonnet 4.6
  • Best for complex tasks — codebases, documents, long workflows
  • Not always needed — use strategically to control costs
  • Available everywhere — Claude Platform, AWS, GCP, Azure

Ready to upgrade?

openclaw config set agents.defaults.contextTokens 1000000