Claude Sonnet 4.6 on VM0. The default agent model
The default for most VM0 agents. Strong tool routing, good long-context behaviour, and the credit baseline. Every other model is priced relative to Sonnet 4.6.
1M tokens · Text / Vision / Code · Prompt cache
Claude Sonnet 4.6 is the workhorse of the Claude 4 family and the default Built-in model on VM0. It picks the right tool with the right arguments more reliably than anything cheaper, stays coherent across hundred-thousand-token conversations, and most production agents — Slack triage, GitHub PR review, customer support — never need to be promoted past it.
Vendor list price is $3 / $15 per 1M tokens, with cached input dropping to $0.30 / 1M. Reach for Opus only when Sonnet visibly fails on the hardest reasoning, and for Haiku or DeepSeek V4 Flash when unit cost dominates.
What is Claude Sonnet 4.6?
February 2026 (Claude 4.6 generation) · Mid-tier of the Claude 4 family. Anthropic's workhorse model, sitting between Haiku and Opus.
Claude Sonnet 4.6 sits in the middle of Anthropic's Claude 4 family. It is the workhorse model designed to handle the full breadth of typical agent work. Multi-tool routing, code edits, long-running conversations, and structured-output tasks. Without the cost premium of Opus.
Across VM0's Built-in lineup, every other model's credit multiplier is normalised against Sonnet 4.6 (×1). That makes Sonnet the right pick when you want predictable budget conversations: “this agent runs at roughly 2× a Sonnet step” is a more useful sentence than absolute dollar quotes that move every quarter.
Sonnet 4.6 supports Anthropic's prompt caching, which makes a big difference for VM0 agents that ship a stable system prompt and a fixed tool schema. Cached input tokens bill at $0.30 per 1M instead of $3. A 10× saving on the parts of the prompt that don't change between turns.
What's notable about Claude Sonnet 4.6
Headline architecture and capability features.
Sonnet 4.6 ships with the 1M-token context window at standard pricing, adaptive thinking inherited from Opus 4.6, and prompt caching that bills cached input at one-tenth the input rate. It accepts multimodal input across text, vision, and code.
Specs at a glance
Claude Sonnet 4.6 benchmarks
Sonnet 4.6 sits roughly 3 to 4 percentage points behind Opus 4.6 on Anthropic's headline coding benchmarks while being three to five times cheaper at the vendor level. The typical Opus/Sonnet trade-off.
Claude Sonnet 4.6 pricing
Provider list price, per 1M tokens.
How Claude Sonnet 4.6 behaves in practice
Observed behaviour from production agent runs.
Tool routing
Best-in-class tool-routing accuracy at this price. On multi-tool flows across Slack, GitHub, Linear, and Notion, Sonnet 4.6 picks the correct tool with the correct arguments more reliably than any model below ×1.7.
Long-context coherence
Coherent across 100K+ token transcripts. Drops below Opus 4.7 only on the longest, most adversarial runs.
Speed
Faster than Opus and slower than Haiku. The right speed/quality balance for production agents.
Cost predictability
Pricing is the credit baseline; prompt caching makes the on-VM0 cost especially predictable for agents with fixed system prompts.
Best agent tasks for Claude Sonnet 4.6
The Slack agent that knows where things live
Triages incoming questions, follows up on stalled threads, posts status updates, and answers search-style queries ("who's owning the auth refactor?"). Sonnet's tool-routing accuracy means the right tool gets called with the right arguments on the first try, even when the request is ambiguous, so the agent feels reliable instead of flaky.
The PR review agent that doesn't drown in noise
Sonnet handles the bulk of code-aware work — PR review, test scaffolding, refactor suggestions, bug bisection — without leaving stylistic comments that nobody asked for. The 1M-token context window lets it pull in the related files and prior reviews when it matters, and you only escalate to Opus 4.7 for the patches Sonnet visibly struggles with.
The research agent that makes 20 tool calls in a row
GitHub plus Linear plus Notion plus the web, stitched together across twenty-plus tool turns to answer a question like "why did this customer churn last quarter?" Sonnet keeps the goal in view across the whole chain at a fraction of Opus's cost, which is what makes it sustainable for everyday research as opposed to one-off deep dives.
The customer-support assistant with a stable system prompt
Long conversation histories, frequent tool calls into the CRM, the same hefty system prompt and tool schema on every turn. Sonnet's prompt caching turns that fixed prefix into a fraction of the input cost after the first call, which is what keeps per-conversation cost flat as volume grows.
When to skip Claude Sonnet 4.6
Skip Sonnet 4.6 on the hardest reasoning steps where it visibly drops instructions and you should escalate to Opus 4.7, on bulk classification at high volume where DeepSeek V4 Flash is roughly 50× cheaper, and on latency-critical micro-replies where Haiku 4.5 is meaningfully faster.
Claude Sonnet 4.6 vs other models
Claude Sonnet 4.6 vs Claude Opus 4.7
Sonnet 4.6 is ×1; Opus 4.7 is ×1.7. Sonnet handles most agents; Opus is the upgrade when reasoning depth matters more than throughput. Many teams use Opus as the planner and Sonnet as the worker.
Claude Sonnet 4.6 vs Claude Haiku 4.5
Haiku 4.5 is ×0.3. Three times cheaper than Sonnet. Sonnet leads on tool-routing accuracy and long-context coherence; Haiku wins on speed and cost. Use Haiku as a sub-agent or for high-volume simple flows.
Claude Sonnet 4.6 vs DeepSeek V4 Pro
DeepSeek V4 Pro (×0.3) matches Sonnet on coding benchmarks (vendor-reported SWE-bench Verified) at much lower cost. The trade is some tool-routing reliability and a less-mature safety profile.
Bottom line: should you use Claude Sonnet 4.6?
Start here. Migrate up to Opus 4.7 or down to Haiku 4.5 / DeepSeek V4 Pro once you've seen real production behaviour and know which direction makes sense.
Frequently asked questions
Why is Sonnet 4.6 the default model on VM0 Managed?
It hits the best balance of reasoning quality, tool-routing accuracy, and cost in our lineup. New agents almost always work on Sonnet without further tuning.
What is Claude Sonnet 4.6's context window?
1 million tokens with up to 64K tokens of output per response.
Does Sonnet 4.6 support image input?
Yes. It's multimodal. Text, code, and images.
When should I switch off Sonnet 4.6?
Switch to Opus 4.7 if Sonnet visibly drops the goal on long agent loops or fails on hard code edits. Switch to Haiku 4.5 or DeepSeek V4 Flash for high-volume simple flows where cost dominates.
Is Sonnet 4.6 the same as Sonnet 4.5?
No. 4.6 is the newer generation in the Claude 4 family with better long-context behaviour and adaptive thinking. The vendor pricing per token is identical.
Alternatives
Using Claude Sonnet 4.6 on VM0
Two ways to access Claude Sonnet 4.6 on VM0
VM0 supports Claude Sonnet 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 Sonnet 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 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 Sonnet 4.6 bills at ×1 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 Sonnet 4.6 sits at the ×1 baseline that every other Built-in model is priced against, so it's the unit you compare costs in when picking between models on VM0.
Available on VM0 since Available since launch.