Anthropic just dropped Opus 4.8... (WOAH)

TL;DR

• 1

  Opus 4.8 posts a big coding jump without a price hike — Anthropic claims 69.2% on SWE-bench Pro, roughly 5 points above Opus 4.7 just six weeks later, while keeping standard pricing at $5 per million input tokens and $25 per million output tokens.

• 2

  Fast mode got dramatically more attractive — Berman highlights that Claude’s fast mode remains his favorite and says Opus 4.8 is about 2.5x faster, while the premium dropped enough that it now looks more like 2x the cost instead of the much steeper surcharge before.

• 3

  Dynamic workflows is the real headline feature — Claude Code can now plan work, spin up tens to hundreds of parallel sub-agents, and run adversarial verification, which Berman describes as exactly the kind of capability Anthropic probably couldn’t afford to release until it had more compute.

• 4

  The benchmark story is messy because the “vibe check” disagrees — Berman compares SWE-bench Pro to DeepU and notes that many developers feel GPT-5.5 is stronger in real coding despite Opus 4.7 previously scoring much higher on SWE-bench, suggesting some leaderboard results no longer match lived experience.

• 5

  GPT-5.5 still looks strongest for terminal-style coding — On Terminal-Bench 2.1, Berman points out GPT-5.5 at 78.2 remains clearly ahead of Opus 4.8, which may explain why so many coders still prefer OpenAI for practical, repo-level work.

• 6

  Anthropic’s compute situation appears to have changed everything — Berman ties the launch to Anthropic’s xAI Colossus access and cloud partnerships with Amazon, Bedrock, Vertex AI, and Microsoft Foundry, arguing the company now has enough supply to push usage-heavy features instead of raising token prices.