94,000 Trials Later, the Same Models Cost About 2× Under a Different Harness
On October 13, 2026 the Strands Agents team published about 94,000 Harbor trials. On matched Anthropic runs, estimated cost ran from about $0.74 to $1.57 per trial.
Tagged
On October 13, 2026 the Strands Agents team published about 94,000 Harbor trials. On matched Anthropic runs, estimated cost ran from about $0.74 to $1.57 per trial.
Checked 16 Oct 2026: Qwen Decider 2B v1 scores 180 of 231. Gemma 4 12B, released 15 Oct, scores 200 of 231 at a 136 ms H100 median. Jev stays $0.042/MTok, with no matching JevBench row.
Checked 15 October 2026. The shared MCP adapter exposes 3 tools. execute-ability can run all 7 canonical abilities, including product-delete. Week one allows two reads.
OpenAI's 11 February 2026 Codex experiment merged about 1,500 pull requests. AgentCore Harness, GA 17 June 2026, still defaults to 75 iterations and a 3,600-second timeout — the harness sets those bounds, not the model.
Two CLIs, not one binary named bedrock-cli. AWS CLI 2.37.12 and agentcore 0.30.0, checked 11 October 2026.
Checked 8 October 2026. Storefront MCP is beta, with 7 B2C tools. search_products rejects a term under 3 characters. create_checkout_url takes 0 payment fields and returns a URL.
Checked 8 October 2026. Odoo 19 JSON-2 keys last at most 3 months. saas-19.4 MCP exposes 5 tools by default. action_confirm stays off the week-one agent.
Checked 6 October 2026. Catalog and cart tools on /api/mcp are removed. complete_checkout is off for both non-token tiers. Call UCP 2026-08-25 through an adapter, not an Admin token.
On September 29, 2026 AWS opened Bedrock Managed Agents, powered by OpenAI, in preview: 3 US Regions, no extra preview charge, and an 8-hour compute window in the AgentCore example. AWS runs the conversation. Your compute runs the tools.
Ten real lookups and three must-escalate cases, re-run before every release. A green latency chart is not correctness. Below 16 out of 30 readiness, do not add a write while you tune the set.
Three identities, not one token. The workload role, the shopper or associate, and the downstream API account. A shared admin credential fails the split. AgentCore Identity does not replace your order-system scopes.
Call the REST Management API with X-Auth-Token and a store hash. Webhook destinations must be port 443. One API account should not hold both read-orders and modify-orders for a week-one agent.
We use cookies and similar technologies to analyze site traffic, personalize content, and provide social media features. By clicking “Accept,” you consent to our use of cookies. You can adjust your preferences at any time.