2026-05-02 — views · 3 models

Summarize 1000-word post into 3 bullets — frontier head-to-head

Prompt

Summarize this 1000-word blog post about Anthropic's Claude 4.7 release into exactly 3 bullets, each ≤ 25 words. Return only the bullets, no preamble.

Notes

Same prompt, same input. Latency measured from API call start to last byte. Cost from official $/Mtok pricing (May 2026).

Results — 3 models

claude-opus-4-7 WIN · 1240ms · in 1280 · out 78 · $0.022

• Anthropic shipped Claude 4.7 with 1M-token context now extended to the Opus tier.
• Tool-use determinism improved on high-fan-out parallel calls; cache TTL stays at 5 min.
• Early users report ~12% p95 latency drop on Sonnet/Haiku — Opus latency improvement claimed but unverified.

gpt-5 TIE · 980ms · in 1280 · out 84 · $0.018

• Claude 4.7 released by Anthropic, headline: 1M context arrives on the Opus tier.
• Anthropic claims ~12% p95 latency improvement across the model family.
• Tool-call behavior more deterministic in parallel/high-fan-out scenarios.

gemini-3-pro LOSS · 1120ms · in 1280 · out 102 · $0.014

• Anthropic released Claude 4.7 today with several updates.
• The 1M context window is now available on Opus tier in addition to Sonnet and Haiku, which is a major change for long-context agentic workflows.
• Latency was reduced by approximately 12% according to Anthropic, and tool-use behavior was improved.