Claude Opus 4.6 on VM0
Anthropic's previous flagship. Same multiplier and 1M context as Opus 4.7. Keep it pinned only when an agent has been validated on this exact version.
1M tokens · Text / Vision / Code · Prompt cache
Claude Opus 4.6 was Anthropic's flagship before Opus 4.7 and introduced most of what now defines the Claude 4 family: the 1M-token context window in beta, adaptive thinking at four effort levels, and the highest agentic-coding scores Anthropic had shipped at the time (vendor-reported SWE-bench Verified 80.8%, Terminal-Bench 2.0 65.4%, OSWorld 72.7%).
Vendor list price is the same $5 / $25 per 1M tokens as 4.7. The only good reason to stay on 4.6 is behaviour stability for an agent that's already been validated against this version; anything new should start on 4.7.
What is Claude Opus 4.6?
February 5, 2026 · Previous top-tier of the Claude 4 family. Superseded by Claude Opus 4.7.
Claude Opus 4.6 was Anthropic's frontier model before Opus 4.7. It was released on February 5, 2026 and introduced several capabilities that defined the Claude 4 family. Adaptive thinking with four effort levels, the 1M-token context window in beta, and Anthropic's highest agentic-coding scores at release.
On VM0 it sits at the same ×1.7 credit multiplier as Opus 4.7. Anthropic explicitly recommends migrating to 4.7 for new work; pin 4.6 only if a specific agent has been validated against this version and you don't want to re-run regression tests yet.
What's notable about Claude Opus 4.6
Headline architecture and capability features.
Opus 4.6 introduced adaptive thinking with four effort levels (low, medium, high, and max, with high as the default) and the 1M-token context window in beta at standard pricing. It added a Compaction API for server-side context summarisation, disabled prefilling as a breaking change versus Opus 4.5 (use structured outputs instead), and shipped a Mailbox Protocol for multi-agent peer-to-peer teams. An inference_geo parameter exposes US-only inference at a 1.1× multiplier.
Specs at a glance
Claude Opus 4.6 benchmarks
Vendor-reported scores from Anthropic's Opus 4.6 release materials and Artificial Analysis. Treat absolute SWE-bench numbers cautiously. OpenAI flagged training-data contamination on SWE-bench Verified across all frontier models.
Claude Opus 4.6 pricing
Provider list price, per 1M tokens.
How Claude Opus 4.6 behaves in practice
Observed behaviour from production agent runs.
Reasoning
Strong on hard reasoning steps. Opus 4.7 is incrementally better at slightly lower vendor cost. There is no benchmark category where 4.6 leads.
Tool use
Reliable across multi-tool agent flows. Same ballpark as Sonnet 4.6 on routing accuracy with extra robustness on edge cases.
Long context
1M-token context with 76% MRCR v2 recall. Actually usable across the full window, not just nominal.
Speed
Slower than Sonnet 4.6 and Haiku 4.5; comparable to Opus 4.7. Around 41 tokens/sec at max effort per Artificial Analysis.
Best agent tasks for Claude Opus 4.6
The production agent that's already paying its way
Your team spent two weeks tuning prompts and tool schemas against Opus 4.6, the agent has been live for a month, and customers are happy. Pinning to 4.6 keeps the behaviour identical while you decide whether the 4.7 upgrade is worth a re-validation cycle, instead of letting Anthropic auto-upgrade your traffic and quietly shifting outputs underneath you.
The regression baseline for an Opus 4.7 rollout
Run the same prompt set through 4.6 and 4.7 side by side, diff the outputs, and decide where the upgrade actually changes behaviour before you flip the switch in production. Same vendor price, same multiplier, identical interface — the only thing different is the model weights, which is exactly what you want when you're isolating regressions.
When to skip Claude Opus 4.6
Don't start new agents on Opus 4.6 unless you have a concrete reason, since 4.7 ships at the same multiplier with stronger behaviour and a lower vendor list price. Anything cost-sensitive should go to 4.7 for the same reason.
Claude Opus 4.6 vs other models
Claude Opus 4.6 vs Claude Opus 4.7
Same ×1.7 multiplier and 1M context window. Opus 4.7 is newer, faster, and lower vendor list price. Pin 4.6 only when you've already invested in tuning against this version.
Claude Opus 4.6 vs Claude Sonnet 4.6
Sonnet 4.6 is ×1 and handles most agent loops. Reach for Opus only when Sonnet visibly fails. Usually for orchestration or hard code edits.
Claude Opus 4.6 vs Kimi K2.6
Kimi K2.6 (×0.3) edges Opus 4.6 on SWE-bench Pro (58.6 vs 53.4 vendor-reported) and is much cheaper. Opus 4.6 retains the safety-profile advantage and is the default Western enterprise pick.
Bottom line: should you use Claude Opus 4.6?
Pin if you've already validated against it; otherwise start on Opus 4.7. The migration is a setting change, not a rewrite.
Frequently asked questions
When was Claude Opus 4.6 released?
Anthropic released Opus 4.6 on February 5, 2026. Opus 4.7 followed shortly after.
Should I migrate from Opus 4.6 to Opus 4.7?
Yes for new work. Same multiplier, same 1M context, lower vendor list price, stronger behaviour on agentic-coding tasks. Migrate pinned agents only after running them through your regression suite.
What is Claude Opus 4.6's context window?
1 million tokens (beta) with up to 128K tokens of output per response.
Why is Opus 4.6 the default on the Anthropic API key provider?
Historical default from before Opus 4.7 launched. You can switch any agent to Opus 4.7, Sonnet 4.6, or Haiku 4.5 in VM0 Settings → Model Providers without changing the API key.
What's adaptive thinking?
A scheduling layer that lets Claude decide how much reasoning compute to spend per turn. Four levels. Low, medium, high, max. With high as the default. Replaced Opus 4.5's extended-thinking toggle.
Alternatives
Using Claude Opus 4.6 on VM0
Two ways to access Claude Opus 4.6 on VM0
VM0 supports Claude Opus 4.6 as a Built-in model billed in VM0 credits, and through bring-your-own with a Anthropic API key. The Built-in path uses VM0 Managed routing and the credit multiplier explained below; the bring-your-own path bills you directly with the upstream vendor and skips the VM0 credit conversion entirely.
VM0's recommendation
VM0 positions Claude Opus 4.6 as a core agent model, recommended alongside Claude Opus 4.7, Claude Opus 4.6, and Claude Sonnet 4.6 for the steps that drive the actual outcome of an agent run. These are the models we'd pick for the orchestrator role, for code-touching agents, and for any step where a wrong answer is expensive.
Credits and the ×1.7 multiplier
Every Built-in model on VM0 is priced as a multiple of Claude Sonnet 4.6, which sits at the ×1 credit baseline. Claude Opus 4.6 bills at ×1.7 credits. The multiplier is what shows up on your VM0 invoice; the vendor list price in the pricing table above is what the upstream provider charges before VM0 converts it into credits.
Claude Opus 4.6 bills at ×1.7, which means a step here costs 1.7× the credits of an equivalent step on Sonnet 4.6 (the ×1 baseline). It's a premium tier on VM0, so the cost-effective pattern is to default to a cheaper model and route only the steps that genuinely need the extra reasoning depth to Claude Opus 4.6.
Available on VM0 since Available since launch.