evals.report
BenchmarksLabsCompareRun guidesIn the wild
AnthropicClaude Opus

Claude Opus 4.8

Anthropic · Claude Opus. Released May 28, 2026.

Claude Opus 4.8 is a model from Anthropic in the Claude Opus family, released May 28, 2026. evals.report tracks 30 reported Claude Opus 4.8 benchmark scores across SWE-bench Verified, GPQA Diamond, Humanity's Last Exam, DeepSWE, ARC-AGI-1, ARC-AGI-3, ARC-AGI-2, LiveBench, and 22 more — each shown with its benchmark, metric, source status, and date, and never combined into a single ranking.

30 results

Benchmark results 30

Compare this model
BenchmarkCategoryScoreMetricStatusDate
SWE-bench VerifiedCoding88.6%% resolvedVerifiedMay 28, 2026Details
GPQA DiamondReasoning93.6%accuracyVerifiedMay 28, 2026Details
Humanity's Last ExamReasoning49.8%accuracyVerifiedMay 28, 2026Details
DeepSWECoding58%% resolvedVerifiedMay 28, 2026Details
ARC-AGI-1Reasoning92%accuracyOfficialMay 28, 2026Details
ARC-AGI-3Reasoning1.52%accuracyOfficialMay 28, 2026Details
ARC-AGI-2Reasoning72.08%accuracyOfficialMay 28, 2026Details
LiveBenchReasoning77.22%scoreOfficialMay 28, 2026Details
SWE-bench ProCoding69.2%% resolvedUnverifiedMay 28, 2026Details
GBA EvalCoding70.9%overall scoreOfficialMay 28, 2026Details
WeirdMLCoding82.9%average accuracyOfficialMay 28, 2026Details
CursorBenchAgents63.8%scoreOfficialMay 28, 2026Details
PostTrainBenchAgents37.23%weighted average scoreOfficialMay 28, 2026Details
Artificial Analysis Intelligence IndexReasoning61IndexOfficialMay 28, 2026Details
OSWorldAgents83.4%task success rateVerifiedMay 28, 2026Details
τ²-bench (Telecom)Tool use94.4%pass^1OfficialMay 28, 2026Details
AIME 2026Reasoning100.00%accuracyOfficialMay 28, 2026Details
GDPvalAgents1890EloOfficialMay 28, 2026Details
AA-Omniscience: Knowledge and Hallucination BenchmarkReasoning27AA-Omniscience IndexOfficialMay 28, 2026Details
IFBenchReasoning62.2%accuracyOfficialMay 28, 2026Details
EQ-Bench Creative Writing v3Chat preference2031EloVerifiedMay 28, 2026Details
Design ArenaChat preference1282EloVerifiedMay 28, 2026Details
CharXivMultimodal89.9%accuracyUnverifiedMay 28, 2026Details
ScreenSpot-ProMultimodal87.9%accuracyUnverifiedMay 28, 2026Details
SWE-bench MultimodalCoding38.4%% resolvedVerifiedMay 28, 2026Details
MathArena HMMT February 2026Reasoning95.45%accuracyOfficialMay 28, 2026Details
Vibe Code BenchCoding82.72%Overall accuracyOfficialMay 28, 2026Details
Remote Labor IndexAgents8.33%automation rateOfficialMay 28, 2026Details
FrontierSWEAgents75%dominance scoreOfficialMay 28, 2026Details
FrontierCodeCoding13.4%weighted score (Diamond)OfficialMay 28, 2026Details

In the wild 3

See all

Real-world feedback on Claude Opus 4.8 from people using it on actual prompts — praise and criticism alike, each linked to its source. Qualitative, never scored.

Report tone

Report type

Topic

Prem
X·@btrmasaladosa·
Positive
i was stuck on a landing page redesign with gpt 5.5 and opus 4.6 since a couple of days. gave a fresh try with opus 4.8 and it one shotted what i was looking for

Task A landing-page redesign GPT-5.5 and Opus 4.6 hadn't cracked over several days.

anecdotal
Mark Santos
X·@markksantos·
Mixed
I think in terms of application functionality, quality, and design, Opus won … got stuck twice in a retry loop (had to prompt to self-correct).

Task Head-to-head vs Sakana Fugu Ultra: a single-file Three.js Crossy Road game.

prompt shownsingle run
Machina
X·@EXM7777·
Mixed
Opus 4.8 in the last 48hrs is amazing … it's just very sad to go from godlike performance to barely usable some days.
anecdotal