VerdictPal
Join

Models / Anthropic

Claude Opus 4.8

Claude Opus · Released 2026-05-28

Anthropic's May 2026 flagship. Tops the Vals Index (70.17%) and the Artificial Analysis Intelligence Index (61.4). SWE-bench Verified 88.6%, Terminal-Bench 2.0 74.6%, 1M-token context, $15 / $75 per 1M tokens.

Front-door model for hard reasoning, long documents, agentic coding, and tool-heavy research. The model we reach for first when a task is too hard for Sonnet but small enough to fit in one prompt.

#5 of 21 on Vals Index · current snapshot
70 Vals Index · 26 benchmark rows · 12 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite73
Context1M
Input / 1M$15
Output / 1M$75
Knowledge cutoff2026-02-01
Output speed63 t/s
TTFT (default API)30.13s
Statusflagship
ClaudeFactoryPerplexity
KnowledgeReasoningMathCodingAgenticMultimodalLong contextTool useSafetyHuman preference
Knowledge
91
Reasoning
80
Math
94
Coding
96
Agentic
100
Multimodal
71
Long context
58
Tool use
87
Safety
92
Human preference
92

86average across 10 tested families

Editor's note
Opus 4.8 is the model we reach for first when a task is hard enough to make a Sonnet hallucinate but small enough to fit in one prompt. The 1M context window is real, not marketing — long-document citation fidelity and BFCL scores back it up. The single-number rank (Vals, AA) is the highest we've measured.
Benchmark placements

Where Claude Opus 4.8 places on each public benchmark source that publishes a row. Rank counts every model with a latest row in the same test — not a universal quality score.

BenchmarkFamilyRankScoreSourceDate
LongBench v2Long context#1/ 1958LongBench v2 leaderboard2026-04-01
MMLUKnowledge#1/ 2291Anthropic Claude 4 family2026-05-28
AIMEMath#2/ 1994Anthropic Claude 4 family2026-05-28
LMArena (Chatbot Arena)Human preference#2/ 1392LMArena leaderboard2026-05-25
GDPvalAgentic#2/ 6100Artificial Analysis — GDPval-AA v22026-06-17
HELM SafetySafety#2/ 992Stanford HELM leaderboard2026-05-28
LiveBenchReasoning#2/ 2171LiveBench leaderboard2026-04-15
MCP AtlasTool use#2/ 1079MCP Atlas leaderboard2026-06-09
SWE-bench VerifiedCoding#2/ 2389SWE-bench official leaderboard2026-05-30
BFCLTool use#3/ 2087Berkeley Function Calling Leaderboard V42026-04-15
DeepSWECoding#3/ 659DeepSWE leaderboard v1.12026-06-20
Humanity's Last ExamReasoning#3/ 6546Artificial Analysis2026-07-21
HumanEvalCoding#3/ 1596Anthropic Claude 4 family2026-05-28
MMMUMultimodal#3/ 871MMMU leaderboard2026-05-28
Artificial Analysis Intelligence IndexReasoning#4/ 6756Artificial Analysis2026-07-21
Vals IndexAgentic#5/ 2070Vals AI — Vals Index2026-06-04
MATHMath#6/ 2088Anthropic Claude 4 family2026-05-28
MMLU-ProKnowledge#9/ 2182MMLU-Pro HuggingFace leaderboard2026-05-28
AA time to first tokenPerformance#13/ 69Artificial Analysis2026-07-21
SciCodeCoding#14/ 6554Artificial Analysis2026-07-21
Terminal-BenchAgentic#17/ 6975Terminal-Bench 2.0 leaderboard2026-05-20
τ³-Bench BankingAgentic#30/ 6428Artificial Analysis2026-07-21
IFBenchReasoning#41/ 5762Artificial Analysis2026-07-21
AA output speedPerformance#45/ 69Artificial Analysis2026-07-21
AA-LCRLong context#48/ 6458Artificial Analysis — AA-LCR2026-06-17
GPQA DiamondReasoning#51/ 6980GPQA Diamond paper leaderboard2026-05-15
Family context

The three highest-scoring carded models in each capability family. Where Claude Opus 4.8 shows up, it's highlighted.

KnowledgeClaude Opus 4.8 · 91
  1. 1
    GPT 5.5
    91
  2. 2
    Claude Opus 4.8
    91
  3. 3
    Claude Sonnet 4.6
    91
ReasoningClaude Opus 4.8 · 80
  1. 1
    Claude Mythos Preview
    95
  2. 2
    GPT-5.6 Sol
    94
  3. 3
    Kimi K3
    94
MathClaude Opus 4.8 · 94
  1. 1
    OpenAI o3
    99
  2. 2
    OpenAI o1
    97
  3. 3
    Gemini 3 Pro
    96
CodingClaude Opus 4.8 · 96
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticClaude Opus 4.8 · 100
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
MultimodalClaude Opus 4.8 · 71
  1. 1
    Gemini 3.1 Pro
    72
  2. 2
    GPT 5.5
    72
  3. 3
    Claude Opus 4.8
    71
Long contextClaude Opus 4.8 · 58
  1. 1
    Kimi K3
    75
  2. 2
    GPT-5.1
    75
  3. 3
    GPT-5.6 Sol
    74
Tool useClaude Opus 4.8 · 87
  1. 1
    Claude Sonnet 4.6
    89
  2. 2
    GPT 5.5
    87
  3. 3
    Claude Opus 4.8
    87
SafetyClaude Opus 4.8 · 92
  1. 1
    Claude Opus 4 (2024-07)
    92
  2. 2
    Claude Opus 4.8
    92
  3. 3
    GPT 5.5
    91
Human preferenceClaude Opus 4.8 · 92
  1. 1
    GPT 5.5
    95
  2. 2
    Claude Opus 4.8
    92
  3. 3
    Gemini 3.5 Flash
    91
Newest receipts

Benchmark rows added to the public ledger for Claude Opus 4.8 in the last 120 days. Older rows live in the full table below.

26 recent
BenchmarkFamilyScoreSourceDays ago
AA output speedPerformance63 t/sArtificial Analysis2
AA time to first tokenPerformance30.13sArtificial Analysis2
Artificial Analysis Intelligence IndexReasoning55.7%Artificial Analysis2
Humanity's Last ExamReasoning45.7%Artificial Analysis2
IFBenchReasoning62.24%Artificial Analysis2
SciCodeCoding53.5%Artificial Analysis2
τ³-Bench BankingAgentic27.63%Artificial Analysis2
DeepSWECoding59%DeepSWE leaderboard v1.133
GDPvalAgentic1638Artificial Analysis — GDPval-AA v236
AA-LCRLong context58.3%Artificial Analysis — AA-LCR36
MCP AtlasTool use79.2%MCP Atlas leaderboard44
Vals IndexAgentic70.4%Vals AI — Vals Index49
SWE-bench VerifiedCoding88.6%SWE-bench official leaderboard54
AIMEMath94.2%Anthropic Claude 4 family56
HELM SafetySafety91.8%Stanford HELM leaderboard56
HumanEvalCoding95.8% pass@1Anthropic Claude 4 family56
MATHMath88.4%Anthropic Claude 4 family56
MMLU-ProKnowledge82.4%MMLU-Pro HuggingFace leaderboard56
MMLUKnowledge91.4%Anthropic Claude 4 family56
MMMUMultimodal70.6%MMMU leaderboard56
LMArena (Chatbot Arena)Human preference1462LMArena leaderboard59
Terminal-BenchAgentic74.6%Terminal-Bench 2.0 leaderboard64
GPQA DiamondReasoning79.8%GPQA Diamond paper leaderboard69
BFCLTool use86.9%Berkeley Function Calling Leaderboard V499
LiveBenchReasoning70.8%LiveBench leaderboard99
LongBench v2Long context58.4%LongBench v2 leaderboard113
Family coverage

A high composite that hides a weak family is a trap. These bars surface the families where this model hasn't been publicly tested, and where it leads.

Knowledge91
Reasoning80
Math94
Coding96
Agentic100
Multimodal71
Long context58
Tool use87
Safety92
Human preference92
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover or tab any dot for name, score, and input price.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMiniMaxxAIMetaZhipuXiaomiMeituanCohere

Full benchmark ledger

Every catalog benchmark for Claude Opus 4.8. Scores link to the original source; gaps mean no public row exists yet.

26 sourced rows

26 of 32 catalog benchmarks have a sourced row for Claude Opus 4.8.

Knowledge

Reasoning

Math

Coding

Agentic

Multimodal

Long context

Tool use

Performance

Safety

Human preference

Changelog
  1. Released Claude Opus 4.8 — top of Vals (70.17%) and AA (61.4).
  2. Claude Opus 4.7 retired for general availability.
Sources