Descriptions:
Matthew Berman delivers a structured analysis of GPT 5.5 following two weeks of personal pre-release testing, covering benchmark data, pricing, token efficiency improvements, and the strategic context behind OpenAI’s increasingly enterprise-focused model roadmap. The core technical story: GPT 5.5 completes the same tasks as GPT 5.4 with roughly half the output tokens — TerminalBench shows 39.1 at 2,165 output tokens versus 34.2 at 4,950 tokens — meaning the effective cost per task decreases despite the per-token price doubling from $2.50/$15 to $5/$30.
On enterprise benchmarks provided by Box AI, GPT 5.5 shows significant accuracy jumps across industry verticals: healthcare improves from 61% to 78%, financial services sees nearly a 20-point gain, and the full-dataset accuracy index moves from 67% to 77% over GPT 5.4. Berman also notes a meaningful personality improvement — GPT 5.5 responds more concisely and directly than 5.4, reducing the over-explanation friction that made agentic coding sessions in Codeex tedious.
Berman frames GPT 5.5 explicitly within OpenAI’s response to Anthropic’s $30 billion ARR run rate, driven largely by enterprise coding adoption. The self-reinforcing flywheel — better coding model, more enterprise revenue, more training data, better next model — is the strategic logic he argues explains OpenAI’s current investment prioritization. For enterprise developers evaluating whether to shift primary model usage from Claude Opus 4.7 to GPT 5.5, the pricing math and Box AI vertical benchmarks are the most decision-relevant data points in the video.
📺 Source: Matthew Berman · Published April 23, 2026
🏷️ Format: News Analysis
