VerdictPal
Join

Models / Anthropic

Claude Mythos Preview

Claude Opus · Released 2026-06-01

Anthropic's June 2026 research preview. Tops SWE-bench Verified (93.9%, best-of-3 harness) and the Vals Index (73.42%). Also leads GPQA Diamond (94.6%), HLE (64.7%), and CharXiv (93.2%). 2M context, $75 / $300 per 1M tokens.

Research preview for the hardest reasoning, coding, and generalization tasks. Not yet in general availability — access via the Claude research program. Treat as a research artifact, not a deployment candidate.

#2 of 24 on Vals Index · current snapshot
73 Vals Index · 5 benchmark rows · 4 sourcesPosition is benchmark-specific — not a cross-family or cross-source ranking.
Composite73
Context2M
Input / 1M$75
Output / 1M$300
Knowledge cutoff2026-04-01
Statusflagship

Family profile

Best published score in each covered benchmark family.

91/100 avg
Reasoning
95
Coding
94
Agentic
84

3 tested benchmark families

Editor's note
Treat Mythos as a research artifact, not a deployment candidate. The SWE-bench 93.9% uses a best-of-3 agent harness — production single-attempt scores are materially lower. The Vals 73.42% is the highest number on the public leaderboard, but the score is on a research version with frontier inference cost.
Benchmark placements

Where Claude Mythos Preview 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
SWE-bench VerifiedCoding#1/ 2394SWE-bench official leaderboard2026-06-02
GPQA DiamondReasoning#2/ 18595GPQA Diamond paper leaderboard2026-06-02
Vals IndexAgentic#2/ 2373Vals AI — Vals Index2026-06-04
ARC-AGIReasoning#12/ 1424ARC Prize leaderboard2026-06-02
Terminal-BenchAgentic#13/ 18484Terminal-Bench 2.0 leaderboard2026-06-02
Family context

The three highest-scoring models with pages in each capability family. Where Claude Mythos Preview shows up, it's highlighted.

ReasoningClaude Mythos Preview · 95
  1. 1
    Claude Mythos Preview
    95
  2. 2
    Gemini 3.7 Flash
    95
  3. 3
    Grok 4.6
    95
CodingClaude Mythos Preview · 94
  1. 1
    GPT 5.5
    96
  2. 2
    Claude Opus 4.8
    96
  3. 3
    Claude Opus 4.6
    96
AgenticClaude Mythos Preview · 84
  1. 1
    GPT 5.5
    100
  2. 2
    Claude Opus 4.8
    100
  3. 3
    Claude Fable 5
    100
Newest receipts

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

5 recent
BenchmarkFamilyScoreSourceDays ago
Vals IndexAgentic73.42%Vals AI — Vals Index94
SWE-bench VerifiedCoding93.9%SWE-bench official leaderboard96
Terminal-BenchAgentic84.1%Terminal-Bench 2.0 leaderboard96
ARC-AGIReasoning24.1%ARC Prize leaderboard96
GPQA DiamondReasoning94.6%GPQA Diamond paper leaderboard96
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.

Reasoning95
Coding94
Agentic84
Price vs. performance
0255075100$0.10$0.30$1$3$10$30input USD / 1M tokens · log scale →← intelligence

Hover any dot for name, score, and input price. Keyboard: tab through the top twelve, or use the ranking below.

AnthropicOpenAIMistralAlibabaDeepSeekGoogleMoonshotMetaMiniMaxxAIZhipuXiaomiOtherMeituanCohereNVIDIA

Full benchmark ledger

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

5 sourced rows

5 of 37 catalog benchmarks have a sourced row for Claude Mythos Preview.

Reasoning

Coding

Agentic

Changelog
  1. Released Claude Mythos research preview — tops Vals (73.42%) and SWE-bench Verified (93.9%, best-of-3).
Sources