Anthropic · May 27, 2026
Should you upgrade? Yes — and here's the proof in three numbers: 69.2% SWE-Bench Pro, 1,890 Arena Elo, $5/$25 (unchanged).
SWE-Bench Pro69.2%Arena Elo1,890API pricing$5 / $25Parallel subagentsNative ✓
SWE-Bench Pro64.3%Arena Elo~1,800API pricing$5 / $25Parallel subagentsManual only
From official Anthropic release data · May 27, 2026
Three real improvements — not just a version bump.
Spawn, coordinate, and merge multiple agents natively. No custom orchestration needed.
69.2% — the highest public coding score of any released model. Up 4.9 pts from 4.7.
121 points ahead of GPT-5.5. Implies a ~67% win rate in head-to-head preference testing.
API pricing — both models
Upgrade cost
$0
4.8 also adds Fast mode at $10/$50 — 3× cheaper than 4.7's Fast mode ($30/$150 per million).
Those are the only valid reasons to wait.
One line. That's it.
The API interface is identical. Anthropic API docs →
Related
