{"slug": "amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets", "title": "Amazon Replaces Claude Sonnet 4.6 With Strands Decider on Internal Tickets", "summary": "Amazon engineers replaced Claude Sonnet 4.6 with Strands Decider E4B for labeling internal support tickets, cutting modeled cost from about $22,200 to $500 per million tickets at published list prices, according to a November 9, 2026 write-up by the Strands Agents team. The Strands Decider E4B, a LoRA adapter and readout head on Google's Gemma 4 E4B served on SageMaker ml.g5.xlarge, cut average latency from 4,215 ms to 897 ms and input tokens per ticket from 6,554 to 2,009, while Sonnet 4.6 remains on Bedrock global endpoints at $3/$15 per million input/output tokens. Amazon ran a DUAL mode with Sonnet labels authoritative before switching full traffic to the small model, and the team has not independently rerun Amazon's quality evaluators, so accuracy claims remain Amazon's report.", "body_md": "**Amazon engineers replaced Claude Sonnet 4.6 with Strands Decider E4B** for labeling internal support tickets — cutting modeled cost from about **$22,200 to $500 per million tickets** at published list prices, according to the [Strands Agents team write-up](https://strandsagents.com/blog/amazon-saves-millions-replacing-sonnet-with-strands-decider/?ref=explainx) dated November 9, 2026 that reached the [Hacker News front page](https://news.ycombinator.com/item?id=50436640&ref=explainx) the following day.**\n\nThis is the clearest public teardown yet of **when to fire a frontier LLM** and ship a **tiny decision head** instead.\n\n## TL;DR\n\n| Metric | Sonnet 4.6 (reported) | Strands Decider E4B | \n|---|---|---|\n| Avg latency | 4,215 ms | 897 ms | \n| Input tokens / ticket | 6,554 | 2,009 | \n| Output tokens / ticket | 170 | 1 | \n| Cost / million tickets (list) | ~$22,200 | ~$500 | \n| Task | Multi-class ticket labels | Same fixed catalog | \n\n## Problem shape: closed catalog at huge volume\n\nAmazon Strands Decider cut internal ticket labeling cost versus Claude Sonnet 4.6\n\nInternal ticketing systems classify each ticket into **known categories** — routing, prioritization, SLA rules. That is structurally **multi-class classification**, not open-ended chat.\n\nSonnet 4.6 worked when Amazon needed a quick solution: strong JSON obedience, decent reasoning, available on Bedrock. But every ticket paid for **thousands of input tokens** and **hundreds of output tokens** because the model generates language, not just a label index.\n\nStrands Decider flips the contract:\n\n- Inputs describe ticket state.\n- Outputs are **scores over a fixed catalog** (about 20 categories in the blog's example).\n- No marketing prose, no chain-of-thought essay — **one decision** .\n\nThat's the same design pattern we teach for [semantic caching and routing](https://explainx.ai/blog/semantic-caching-llm-apps-cost-latency-guide-2026), taken to its logical extreme.\n\n## Rollout: DUAL mode before cutover\n\nAmazon did not blind-switch. The blog describes modes:\n\n- **DUAL** — Sonnet and Decider run in parallel; Sonnet labels remain authoritative while teams compare.\n- **DECIDER** — full traffic on the small model after parity looked good.\n\nThat is the enterprise version of **shadow deployments** in traditional ML. Agent teams should copy it before any [Bedrock tokenomics refactor](https://explainx.ai/blog/amazon-bedrock-tokenomics-data-lake-finops-2026).\n\n## Infrastructure and pricing math\n\nDecider E4B (a LoRA adapter and readout head on Google's Gemma 4 E4B, per its [Hugging Face model card](https://huggingface.co/StrandsAgents/strands-decider-E4B-gemma4-v1-2610?ref=explainx)) serves on **SageMaker** `ml.g5.xlarge` (NVIDIA A10G). Sonnet stays on **Bedrock global endpoints** at list **$3 / $15 per million** input/output tokens.\n\nThe blog's table (reproduced conceptually):\n\n|  | Sonnet 4.6 | Decider E4B | \n|---|---|---|\n| Billing model | Per token | Per GPU hour | \n| Per-ticket cost (derived) | ~$0.0222 | ~$0.0005 | \n| Per-million tickets | ~$22,200 | ~$500 | \n\nDecider's per-ticket cost **depends on utilization**. At ~640 ms per request, one instance might handle ~5,600 tickets/hour; the blog assumes headroom for bursts, landing near **$500/million** in practice.\n\nWe have not independently rerun Amazon's quality evaluators. Treat accuracy claims as **Amazon's report**, not a third-party benchmark.\n\n## What developers are saying\n\nOn [HN item 50436640](https://news.ycombinator.com/item?id=50436640&ref=explainx), commenters treated the post as a **production routing** story, not a model-brand fight.\n\n**Kaedon** [asked whether the cited per-million savings imply on the order of tens of millions of internal tickets per year](https://news.ycombinator.com/item?id=50436881&ref=explainx) once alarms and security routing are included — scaling math, not a verified Amazon volume figure.\n\n**Gasp0de** [said the Gemma-based E4B build versus the earlier 2B Strands Decider trial was easy to misread](https://news.ycombinator.com/item?id=50437514&ref=explainx) and wondered why the shipped endpoint is still branded Strands Decider.\n\n**psnehanshu** [questioned why Sonnet was the starting point](https://news.ycombinator.com/item?id=50437307&ref=explainx) instead of a smaller chat model — a fair prompt to audit historical routing choices before copying the pattern.\n\nNone of these threads settle quality parity; they show developers stress-testing **volume, naming, and baseline model choice** when a decision head replaces a frontier LLM.\n\n## When to copy Amazon's pattern\n\nRouting high-volume tickets to a small decision model instead of a frontier LLM\n\n| Signal | Route to Decider-style model | Stay on frontier LLM | \n|---|---|---|\n| Label set | Fixed enum < 100 classes | Free-text replies | \n| Volume | Millions / month | Thousands | \n| Risk | Misroute ticket (recoverable) | Customer-facing prose | \n| Change frequency | Quarterly taxonomy updates | Daily new intents | \n| Explainability | Logit scores suffice | Need cited paragraphs | \n\nPair with [Claude Sonnet cache pricing](https://explainx.ai/blog/claude-sonnet-5-5-cache-read-price-cut-0-10-agentic-cost-math-2026) math — caching helps Sonnet, but cannot beat a 44x structural cost gap when outputs are one token.\n\n## AWS and Anthropic context\n\nAmazon helped build Anthropic, but **finops pressure** is real. Separate outlets reported Amazon hunting cheaper models as contracts shift toward token billing. This blog is evidence Amazon **will ship internal replacements** where economics demand it, regardless of partnership press releases.\n\nFor builders on **Bedrock**, the lesson is to architect **pluggable decision stages** in agent graphs — a pattern [AWS Bedrock AgentCore layers](https://explainx.ai/blog/aws-pluggable-agentic-ai-framework-bedrock-agentcore-seven-layers-2026) already encourage.\n\n## Implementation checklist\n\n1. **Instrument token counts per task type** in your agent — if p95 input is 5k tokens for a label, you are a candidate for specialization.\n2. **Freeze a golden ticket set** with human labels.\n3. **Train or host a decision endpoint** (SageMaker, self-hosted Gemma, or your own linear head on embeddings).\n4. **Run DUAL** until mismatch rate is below your routing SLO.\n5. **Keep Sonnet on the long tail** — novel tickets still benefit from generative reasoning.\n\n## Token math worked example (per the Strands blog)\n\nUsing Amazon's published token splits for **one ticket**:\n\n| Step | Sonnet 4.6 | Decider E4B | \n|---|---|---|\n| Input tokens | 6,554 | 2,009 | \n| Output tokens | 170 | 1 | \n| Bedrock list price components | $3 input / $15 output per M | n/a (GPU hour) | \n\nInput cost Sonnet ≈ 6.554 × $3/M ≈ **$0.0197**. Output ≈ 170 × $15/M ≈ **$0.00255**. Total ≈ **$0.0222** per ticket — matching the blog.\n\nDecider swaps token bills for **amortized GPU seconds** at ~897 ms per call. At ~5,600 tickets/hour/instance, the blog lands near **$500 per million tickets** after burst headroom — **44× cheaper** than Sonnet at list.\n\nYour org's **Enterprise Discount Program** on Bedrock changes the Sonnet side; **Savings Plans** change GPU side. Recompute before pitching leadership.\n\n## Architecture diagram in prose\n\nTickets flow: **feature extractor → Strands Decider endpoint → label choice → router**. An optional **Sonnet shadow path** runs in parallel during DUAL mode before cutover.\n\nSonnet path adds **chain-of-thought tokens** you do not need when the downstream system only consumes an **enum**. Removing free-text generation is the savings lever, not squeezing prompt engineering on the same model family.\n\n## Monitoring after cutover\n\nTrack these weekly:\n\n| Metric | Healthy signal | Investigate if | \n|---|---|---|\n| Label mismatch vs golden set | Under agreed threshold | Spikes after taxonomy edit | \n| p99 latency | Near Decider SLA | Approaches old Sonnet tail | \n| \"Other/unknown\" bucket rate | Flat | Rises (novel ticket types) | \n| Human overrides | Stable | Climbs after cutover | \n| GPU utilization | 60–80% | Sustained peg at 100% | \n\nFeed overrides back into **DUAL** shadow runs before the next retrain.\n\n## Relationship to Strands Agents open tooling\n\nThe post lives on **strandsagents.com**, AWS's agent framework site. Even if you are not Amazon, the pattern is portable:\n\n- Export labeled tickets.\n- Fine-tune or distill a **small classifier** .\n- Host on **SageMaker** or any GPU service.\n- Keep Sonnet on the **escalation queue** only.\n\nSee also [AWS Quick Skills catalog](https://explainx.ai/blog/aws-quick-skills-agents-catalog-build-guide-2026) for how Amazon productizes repeatable agent templates — internal Decider is the extreme version of templating.\n\n## What people are asking\n\n**Is 2B enough?** Amazon moved from a 2B Decider trial to **Gemma-based E4B** when evals demanded it. Size is empirical.\n\n**Can I buy Strands Decider on Marketplace?** The blog documents **Amazon internal** deployment; public SKUs may lag. Watch re:Invent channels.\n\n**Does this hurt Anthropic revenue at Amazon?** Likely on this **one workload**. Bedrock still runs plenty of Sonnet elsewhere in the company.\n\n## FinOps template you can copy\n\n| Field | Sonnet baseline | After Decider | \n|---|---|---|\n| Monthly ticket volume | 1.0M (example) | 1.0M | \n| Cost per million (list) | $22,200 | $500 | \n| Monthly inference spend | $22,200 | $500 | \n| GPU ops overhead | $0 | add SRE hours | \n| Break-even if GPU ops | n/a | still wins above ~10k tickets/mo at these list prices | \n\nPlug your real volumes into the Strands table before presenting to finance. Include **human review hours** saved from faster labels — latency dropped **78%** in their report.\n\n## Security note\n\nTicket bodies contain **PII and credentials**. Moving from Bedrock Sonnet to a self-hosted Decider endpoint shifts **data residency** to your VPC — good for compliance, but it concentrates breach impact. Lock down the **SageMaker endpoint** with IAM, VPC-only access, and logging equivalent to what you required from Anthropic.\n\nAgents that relabel tickets are still **agents**; apply least-privilege tool access even when the model is only two billion parameters.\n\n## Where this does not apply\n\nSkip the Decider pattern when:\n\n- Labels require **free-text customer communication** .\n- Taxonomy changes **weekly** from product launches.\n- Errors are **highly asymmetric** (medical triage, security escalation).\n- You need **citations** in the output for auditors.\n\nThose paths still belong on frontier models or human teams — just route with a **cheap classifier gate** first so Sonnet only sees the ambiguous tail.\n\nSchedule quarterly taxonomy reviews so your enum stays stable enough to justify the specialized endpoint.\n\nPublish a short **runbook** now for on-call when mismatch rates spike so engineers know whether to roll back to DUAL mode or widen the catalog.\n\n## Bottom line\n\nAmazon's Strands Decider rollout is a **production case study** in specialized models beating general LLMs on cost **without** changing the business outcome. If your agent pipeline treats every step as a chat completion, you are overpaying.\n\nRead the [Strands Agents post](https://strandsagents.com/blog/amazon-saves-millions-replacing-sonnet-with-strands-decider/?ref=explainx) for token tables and deployment notes, then audit your highest-volume tool calls.\n\n## Related reading\n\n*Cost figures use list prices cited by Strands Agents on November 10, 2026; your Bedrock discounts and GPU utilization will differ.*", "url": "https://wpnews.pro/news/amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets", "canonical_source": "https://explainx.ai/blog/amazon-strands-decider-internal-ticketing-sonnet-replacement-2026?utm_source=rss&utm_medium=feed&utm_campaign=blog", "published_at": "2026-11-10 00:00:00+00:00", "updated_at": "2026-11-10 19:24:52.147520+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-infrastructure", "ai-tools", "machine-learning"], "entities": ["Amazon", "Claude Sonnet 4.6", "Strands Decider E4B", "Strands Agents", "Google", "Gemma 4 E4B", "SageMaker", "Bedrock"], "also_reported_by": [], "alternates": {"html": "https://wpnews.pro/news/amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets", "markdown": "https://wpnews.pro/news/amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets.md", "text": "https://wpnews.pro/news/amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets.txt", "jsonld": "https://wpnews.pro/news/amazon-replaces-claude-sonnet-4-6-with-strands-decider-on-internal-tickets.jsonld"}}