Claude Opus 4.6 — The Price Didn't Change, Everything Else Did
Anthropic just released Claude Opus 4.6. It's their most capable model ever—and it costs exactly the same as Opus 4.5.
That sentence is the entire story. But the details matter if you're spending money on AI.
The Price History
Anthropic's flagship pricing over the last 18 months:
| Model | Input (per 1M) | Output (per 1M) | Max Output | Release |
|---|---|---|---|---|
| Opus 4.6 | $5.00 | $25.00 | 128K | Feb 2026 |
| Opus 4.5 | $5.00 | $25.00 | 64K | Sep 2025 |
| Opus 4.1 | $15.00 | $75.00 | 32K | May 2025 |
| Opus 4.0 | $15.00 | $75.00 | 32K | Mar 2025 |
That's a 3x price cut from Opus 4.0/4.1 to today—while doubling max output and dramatically improving performance. Anthropic is playing the same game DeepSeek pioneered: better models, same or lower prices.
What You Get For $5/$25
Here's what changed between Opus 4.5 and 4.6 at the same price point:
| Spec | Opus 4.5 | Opus 4.6 | Change |
|---|---|---|---|
| Max output tokens | 64,000 | 128,000 | 2x |
| Context window | 200K | 200K (1M beta) | 5x in beta |
| Long-context retrieval (MRCR v2) | 18.5% | 76% | 4x |
| GDPval-AA (economic tasks) | baseline | +190 Elo | Major leap |
| Adaptive thinking | No | Yes | New |
| Effort controls | No | Yes | New |
The 128K output doubling is the quiet headline. Your maximum output spend per request just went from $1.60 to $3.20—but for agentic coding tasks, the ability to generate longer outputs in a single pass can actually reduce total costs by eliminating multi-turn overhead.
How It Stacks Up Against the Competition
Here's the current frontier model pricing landscape:
| Model | Input | Output | Cached Input | Context | Max Output |
|---|---|---|---|---|---|
| Claude Opus 4.6 | $5.00 | $25.00 | $0.50 | 200K | 128K |
| GPT-5.2 | $1.75 | $14.00 | $0.175 | 400K | 128K |
| OpenAI o3 | $2.00 | $8.00 | $0.50 | 200K | 100K |
| Gemini 2.5 Pro | $1.25 | $10.00 | $0.31 | 1M | 65K |
| Claude Sonnet 4.5 | $3.00 | $15.00 | $0.30 | 200K | 64K |
Opus 4.6 is the most expensive per-token on this list. But tokens aren't the whole story. Anthropic's pitch is that Opus 4.6 solves problems in fewer turns and with less supervision—meaning fewer total tokens for the same task.
On the GDPval-AA benchmark (which measures economically valuable agentic tasks), Opus 4.6 outperforms GPT-5.2 by 144 Elo points. If it takes GPT-5.2 three turns to do what Opus does in one, the math flips fast.
The Full Pricing Breakdown
Standard Pricing
| Tier | Input | Output |
|---|---|---|
| Standard | $5.00 / MTok | $25.00 / MTok |
| Batch (50% off) | $2.50 / MTok | $12.50 / MTok |
Prompt Caching
| Cache Type | Price | vs Base |
|---|---|---|
| 5-minute cache writes | $6.25 / MTok | 1.25x |
| 1-hour cache writes | $10.00 / MTok | 2x |
| Cache hits & refreshes | $0.50 / MTok | 0.1x |
Cache hits at $0.50 are 10x cheaper than standard input. If you're sending the same system prompt repeatedly, caching is effectively mandatory.
Long Context Pricing (1M Beta)
| Token Count | Input | Output |
|---|---|---|
| Up to 200K | $5.00 / MTok | $25.00 / MTok |
| Over 200K | $10.00 / MTok | $37.50 / MTok |
The 1M context window (currently in beta for usage tier 4 organizations) doubles input pricing and adds 50% to output pricing beyond 200K tokens. This stacks with batch discounts and caching.
Adaptive Thinking: The New Cost Variable
Opus 4.6 introduces adaptive thinking—the model dynamically decides when to engage extended reasoning. This is different from manually enabling extended thinking on Sonnet.
Four effort levels control the tradeoff:
| Effort | Speed | Intelligence | Token Usage |
|---|---|---|---|
| Low | Fastest | Good | Minimal thinking |
| Medium | Fast | Better | Moderate thinking |
| High | Moderate | Best | Substantial thinking |
| Max | Slowest | Peak | Maximum thinking |
This matters for your bill. Like OpenAI's reasoning tokens (which we covered in Hidden Reasoning Tokens), adaptive thinking tokens count toward output. The difference: Claude's thinking is visible in <thinking> blocks, so you can see exactly what you're paying for.
Pro tip: Start at medium effort. Only bump to high/max for genuinely hard problems—complex code architecture, multi-step math, scientific reasoning.
Where To Get It: Cross-Provider Pricing
We track Opus 4.6 across 10+ providers. Most match Anthropic's direct pricing:
| Provider | Input | Output | Notes |
|---|---|---|---|
| Poe | $4.30 | $21.00 | Cheapest we've found |
| Anthropic (direct) | $5.00 | $25.00 | Official rate |
| Amazon Bedrock | $5.00 | $25.00 | Multiple regions |
| Google Vertex AI | $5.00 | $25.00 | Passthrough pricing |
| Vercel | $5.00 | $25.00 | Passthrough |
| Cloudflare AI Gateway | $5.00 | $25.00 | Passthrough |
| OpenCode | $5.00 | $25.00 | Passthrough |
| Venice | $6.00 | $30.00 | 20% markup |
| Firmware | Free | Free | Free tier |
| GitHub Copilot | Free | Free | Included in plan |
Most providers pass through Anthropic's pricing at cost. The exceptions: Poe undercuts by ~14%, Venice marks up 20%, and Firmware/GitHub Copilot include it in subscription plans.
Sonnet 3.7 Shuts Down February 19
Buried in the same week: Claude Sonnet 3.7 reaches end-of-life on February 19, 2026. That's two weeks from now.
Sonnet 3.7 was $3/$15. Your migration options:
| Model | Input | Output | vs Sonnet 3.7 |
|---|---|---|---|
| Claude Sonnet 4.5 | $3.00 | $15.00 | Same price, better model |
| Claude Haiku 4.5 | $1.00 | $5.00 | 3x cheaper, lighter |
| Claude Opus 4.6 | $5.00 | $25.00 | 1.7x more, flagship |
Sonnet 4.5 is the natural replacement—same price, better performance. If you were already running Sonnet 3.7 at $3/$15, it's a free upgrade.
But if your Sonnet 3.7 workloads involve complex reasoning or agentic tasks, consider whether the jump to Opus 4.6 at $5/$25 actually saves money through fewer turns and higher first-pass accuracy.
The Bottom Line
Claude Opus 4.6 is the rare upgrade that improves everything without raising prices. The 128K output cap, adaptive thinking, and 1M context beta are genuine capability jumps.
If you're currently using:
- Opus 4.5 → Upgrade immediately. Same price, strictly better.
- Opus 4.1 → You're paying 3x more for a weaker model. Switch now.
- Sonnet 3.7 → You have 13 days. Move to Sonnet 4.5 (same price) or Opus 4.6 (more capable).
- GPT-5.2 → Opus 4.6 costs more per token but may cost less per task. Benchmark your workload.
We track Opus 4.6 pricing across all providers in real time. Compare current prices →
Data sourced from Subquery's database of 2,200+ models across 80+ providers. Prices accurate as of February 5, 2026.