TheCircuit: Why the harness matters more than the model

Published on 28.04.2026

AI & AGENTS

Why the harness matters more than the model

TLDR: Ten days of frontier launches from Anthropic, OpenAI, Alibaba, Google, and DeepSeek. Benchmarks are converging, prices are not, and the +5 point bump on Opus 4.7 over Qwen 3.6 Max can cost up to 5.5x more. The interesting question is no longer which model you use — it is how you drive it.

Why the harness matters more than the model

Last week in AI: the unified ChatGPT super-app takes shape

TLDR: OpenAI shipped ChatGPT Images 2.0, Workspace Agents, and previewed GPT-5.5 — positioned as the reasoning core of an emerging unified super-app spanning chat, coding, and browser agents.

GPT-5.5 preview

Intercom doubled engineering velocity with Claude Code in nine months

TLDR: Intercom published a detailed productivity case study showing engineering velocity doubled across the org in nine months of Claude Code adoption.

Intercom Claude Code productivity case study

DeepSeek V4 and Qwen 3.6 Max narrow the open versus closed gap

TLDR: DeepSeek previewed V4 and Alibaba previewed Qwen 3.6 Max — both claim frontier-class agentic and reasoning gains at significantly lower price points than US peers, narrowing the open and closed performance gap.

DeepSeek V4 preview