What Is Claude Opus 4.8?
Claude Opus 4.8 is Anthropic’s flagship AI model, released May 28, 2026 as a direct upgrade to Opus 4.7. Anthropic describes it as “a modest but tangible improvement” rather than a step-change — and the benchmark numbers back that framing up. It’s the model Anthropic recommends by default for new agentic coding work, sitting above Sonnet 4.6 in capability and below the Mythos-class models (Fable 5, currently suspended under a US export-control directive) in raw power.
Key Features
- Strongest agentic coding in the Claude lineup — leads on SWE-bench Pro and OSWorld-Verified, the two benchmarks that matter most for autonomous coding and computer-use tasks.
- Dynamic workflows — a Claude Code research-preview feature that lets Opus 4.8 plan a large task and fan it out across hundreds of parallel subagents in a single session, aimed at codebase-scale migrations and audits.
- Fast Mode, 3x cheaper — runs at 2.5x the standard speed for $10/$50 per million tokens, down from $30/$150 on Opus 4.7’s fast tier.
- 1M token context window — unchanged from Opus 4.7, with the full 128K max output, billed at standard input pricing across the entire window.
- Major math reasoning jump — scored 96.7% on the USAMO 2026 benchmark, up from 69.3% on Opus 4.7, the largest single-cycle math improvement in Opus history.
- Same-day GitHub Copilot availability — shipped as generally available inside Copilot the day it launched.
Benchmarks
Opus 4.8 leads clearly on the benchmarks that matter most for coding and agentic work: 69.2% on SWE-bench Pro (up from 64.3% on Opus 4.7, and well ahead of GPT-5.5’s 58.6%), and 83.4% on OSWorld-Verified for agentic computer use. It’s not a clean sweep — GPT-5.5 still leads on Terminal-Bench 2.1 using its native Codex CLI harness, though the comparison uses different tooling than the standard harness used for other models. On a like-for-like basis, Opus 4.8 is the stronger coding model overall.
Pricing
Opus 4.8 holds the same rate card as Opus 4.7: $5 per million input tokens, $25 per million output tokens. Anthropic didn’t move the sticker price, but two things changed the real cost math: Fast Mode dropped to $10/$50 (a 3x cut), and prompt caching still applies, with cache reads discounted roughly 90% to about $0.50/M on input. If you’re migrating from Opus 4.7, the same tokenizer is in play, so existing cache entries should generally survive the move — worth validating on your own prompts before assuming continuity.
How It Compares
Against GPT-5.5, the honest framing is workload-dependent: Opus 4.8 wins SWE-bench Pro by a wide margin (10.6 points), while GPT-5.5 wins Terminal-Bench on its own harness. Against Gemini 3.1 Pro, Opus 4.8 leads clearly on both SWE-bench Pro (69.2% vs 54.2%) and OSWorld-Verified. The model it can’t currently be directly compared to is Anthropic’s own Fable 5 — Mythos-class and a tier above Opus on raw capability — since Fable 5 has been suspended worldwide since June 12, 2026 under a US export-control directive. See our Fable 5 alternatives guide for how Opus 4.8 fits as the closest same-vendor fallback.
Is Claude Opus 4.8 Worth It?
If you’re running serious agentic coding, codebase-scale migrations, or any workload where reasoning depth matters more than raw speed, Opus 4.8 is the strongest model currently and reliably available from Anthropic — Fable 5 isn’t an option right now, and Opus 4.8 is the direct upgrade path with no API-breaking changes from 4.7. For routine, high-volume production traffic where cost matters more than peak capability, Claude Sonnet 4.6 remains the better-value workhorse beneath it.
Last updated: June 20, 2026.