Claude Opus 4.8 Holds #1 on Artificial Analysis Three Days In, Deployed Across GitHub, AWS, and Google Cloud

Claude Opus 4.8 didn’t need long to find its platform footing. Three days after Anthropic’s May 28 release, the model holds the number one position on Artificial Analysis’s Intelligence Index, per Artificial Analysis’s independent scoring, 61.4 puts it ahead of GPT-5.5 at 60.2, and it’s already live across the three major enterprise deployment paths: Anthropic’s own API, Amazon Bedrock, and Google Cloud Vertex AI.

The integration into GitHub Copilot is the most consequential deployment for developers. Enterprise teams running Copilot now have access to Opus 4.8 without a separate API contract, and the dynamic workflows feature in Claude Code ships with it. That feature enables large-scale parallel problem solving; Anthropic describes coordinating up to 1,000 parallel subagents in its documentation, though that figure comes from Anthropic’s own materials and hasn’t been independently verified. What the platform breadth means practically: a developer who wants to evaluate Opus 4.8 against real agentic tasks has four different access paths available today.

The early tester signal from Anthropic’s release page is worth noting without over-weighting. One tester reported that the model “has noticeably better judgment… catches its own mistakes, pushes back when a plan isn’t sound.” That’s qualitative, attributed to an unspecified tester quoted by Anthropic. It’s consistent with what the benchmark verification brief from May 29 established about the model’s self-correction framing, but it’s not a substitute for independent testing at production scale.

Self-reported benchmarks. Read carefully. Anthropic’s System Card reports SWE-Bench Pro at 69.2%, SWE-Bench Verified at 88.6%, and HLE at 45.7%, all from Anthropic’s internal evaluation. A 4x reduction in unremarked code bugs versus Opus 4.7 is also vendor-reported. Epoch AI’s independent evaluation is pending. Until that evaluation arrives, the Artificial Analysis Intelligence Index score is the only independent data point in the picture, and it measures a different set of capabilities than coding benchmarks do.

Fast mode pricing is confirmed: Anthropic’s page states it’s “now three times cheaper than it was for previous models.” Specific dollar figures have circulated, $10 per million input tokens, $50 per million output tokens in fast mode, but those figures weren’t confirmed in the accessible page content. The 3x reduction is confirmed; the absolute numbers require verification against current Anthropic pricing documentation. Standard pricing is confirmed as “same price” as Opus 4.7, consistent with the $5 per million input and $25 per million output figures that have been reported, though again those specifics need direct verification.

What to watch

Epoch AI’s evaluation when it publishes. The gap between Anthropic’s self-reported SWE-bench figures and whatever independent evaluation produces is the number that will determine whether Opus 4.8 holds enterprise engineering team adoption or loses ground to models with more validated coding performance data. The Artificial Analysis ranking is real and meaningful, but it measures a different dimension than software task completion.

TJS synthesis: Opus 4.8’s three-day trajectory is strong by any measure: independent index leadership, four enterprise deployment paths, and fast mode now accessible at a cost structure that changes the agentic economics. Don’t migrate production workloads from Opus 4.7 based on self-reported benchmarks alone. Wait for Epoch AI’s evaluation, specifically the coding task results, before treating the SWE-bench figures as your decision basis. The platform breadth is already real enough to warrant a pilot evaluation. The benchmark validation isn’t there yet.