Benchmark profile

Vals-hosted GPQA Diamond mirror (Vals GPQA Diamond mirror)

Vals AI hosted GPQA Diamond view with few-shot and zero-shot chain-of-thought task splits.

Data verified July 17, 2026

How BenchLM shows Vals GPQA Diamond mirror

BenchLM mirrors the public Vals AI Vals GPQA Diamond mirror leaderboard captured from https://www.vals.ai/benchmarks/gpqa and updated by Vals on July 17, 2026. The snapshot preserves overall scores, uncertainty, latency, cost-per-test metadata, and task-level scores where Vals publishes them.

Vals GPQA Diamond mirror is display only on BenchLM. Vals proprietary or Vals-hosted aggregate views are useful context, but BenchLM does not use them as weighted ranking inputs or as a replacement for benchmark-native source records.

123 Vals rows3 task viewspublic datasetTasks: Overall, Few-Shot CoT, Zero-Shot CoTDisplay only

Vals GPQA Diamond mirror on Vals AI Vals methodology Vals home

Vals GPQA Diamond mirror score on Vals GPQA Diamond mirror — July 17, 2026

BenchLM mirrors the published vals gpqa diamond mirror score view for Vals GPQA Diamond mirror. Gemini 3.1 Pro Preview leads the public snapshot at 95.45% , followed by GPT-5.6 Sol (95.20%) and Claude Fable 5 (93.18%). BenchLM does not use these results to rank models overall.

Gemini 3.1 Pro Preview

Google

google/gemini-3.1-pro-preview

95.45%

Overall —

GPT-5.6 Sol

OpenAI

openai/gpt-5.6-sol

95.20%

Overall —

Claude Fable 5

Anthropic

anthropic/claude-fable-5

93.18%

Overall —

123 modelsExternal benchmark mirrorsCurrentDisplay onlyUpdated July 17, 2026

Vals GPQA Diamond mirror score table (123 models)

Score

Gemini 3.1 Pro PreviewGoogle

95.45%

GPT-5.6 SolOpenAI

95.20%

Claude Fable 5Anthropic

93.18%

GPT-5.5OpenAI

93.18%

Kimi K3Moonshot AI

92.93%

Grok 4.5SpaceXAI

92.93%

Gemini 3.5 FlashGoogle

92.68%

MiniMax M3MiniMax

92.68%

Claude Opus 4.8Anthropic

92.42%

GPT-5.6 LunaOpenAI

91.67%

Gemini 3 Pro PreviewGoogle

91.67%

GPT-5.2OpenAI

91.67%

GPT-5.4OpenAI

91.67%

Grok 4.3SpaceXAI

91.41%

Muse Spark 1 1Meta

91.16%

GPT-5.6 TerraOpenAI

90.91%

Qwen3.7 MaxAlibaba

90.15%

Claude Opus 4.7Anthropic

90.15%

Muse SparkMeta

89.65%

Claude Opus 4.6 ThinkingAnthropic

89.65%

DeepSeek V4 ProDeepSeek

89.39%

Kimi K2.6Moonshot AI

89.14%

Claude Sonnet 5Anthropic

88.89%

Grok 4.20 0309 ReasoningSpaceXAI

88.64%

Grok 4 0709SpaceXAI

88.13%

Gemini 3 Flash PreviewGoogle

87.88%

Qwen3.5 Plus ThinkingAlibaba

87.37%

Qwen3.6 PlusAlibaba

87.37%

InklingThinkingmachines

87.12%

GPT-5.1OpenAI

86.62%

MiniMax M2.7MiniMax

86.62%

Nemotron 3 Ultra 550b A55bNvidia

86.11%

Claude Opus 4.5 20251101 ThinkingAnthropic

85.86%

GPT-5OpenAI

85.61%

GLM 5.2Zhipu AI

85.61%

Claude Sonnet 4.6Anthropic

85.61%

Grok 4 Fast ReasoningSpaceXAI

85.35%

Qwen3 MaxAlibaba

84.85%

GLM 5.1Zhipu AI

84.52%

Grok 4.1 Fast ReasoningSpaceXAI

84.34%

O3OpenAI

84.09%

Kimi K2.5 ThinkingMoonshot AI

84.09%

GLM 5 ThinkingZhipu AI

83.33%

GPT-5.4 MiniOpenAI

83.08%

Qwen3.5 FlashAlibaba

82.83%

Mimo V2.5 ProXiaomi

82.58%

MiniMax M2.5MiniMax

82.07%

Claude Sonnet 4.5 20250929 ThinkingAnthropic

81.63%

Gemini 2.5 Flash Preview 09 2025Google

81.57%

Mimo V2.5Xiaomi

81.57%

Gemini 3.1 Flash Lite PreviewGoogle

81.06%

Gemini 2.5 Pro Exp 03 25Google

80.81%

GPT-5 MiniOpenAI

80.30%

DeepSeek V3p2 ThinkingFireworks AI

80.30%

GLM 4.7Zhipu AI

80.05%

Claude Opus 4.5Anthropic

79.55%

Qwen3 MaxAlibaba

79.55%

Grok 3 Mini Fast High ReasoningSpaceXAI

79.29%

GPT Oss 120bFireworks AI

78.54%

MiniMax M2.1MiniMax

78.54%

Kimi K2 ThinkingMoonshot AI

78.54%

Qwen3 Max PreviewAlibaba

77.78%

GPT-5.4 NanoOpenAI

77.53%

Gemini 2.5 Flash Preview 09 2025 ThinkingGoogle

76.52%

Claude Opus 4.1 20250805 ThinkingAnthropic

76.26%

DeepSeek V3p2Fireworks AI

76.26%

O3 MiniOpenAI

75.50%

Claude 3.7 Sonnet 20250219 ThinkingAnthropic

75.25%

Claude Sonnet 4 20250514 ThinkingAnthropic

75.00%

O4 MiniOpenAI

74.50%

GLM 4.6Zhipu AI

74.50%

Grok 3SpaceXAI

74.24%

O1OpenAI

73.23%

Grok 3 Mini Fast Low ReasoningSpaceXAI

72.98%

Claude Haiku 4.5 20251001 ThinkingAnthropic

72.22%

GLM 4.5Zhipu AI

72.22%

Claude Opus 4Anthropic

71.72%

Moonshotai Kimi K2 InstructTogether AI

71.46%

Gemini 2.5 Flash Lite Preview 09 2025 ThinkingGoogle

70.20%

Qwen3 235b A22bFireworks AI

70.20%

Claude Opus 4.1Anthropic

69.95%

Llama4 Maverick Instruct BasicFireworks AI

69.44%

Claude Sonnet 4Anthropic

69.44%

GPT Oss 20bFireworks AI

68.94%

Mistral Large 2512Mistral AI

68.43%

GPT-4.1 MiniOpenAI

67.93%

Claude 3.7 SonnetAnthropic

67.42%

GPT-4.1OpenAI

65.40%

Gemini 2.0 Flash 001Google

65.15%

Grok 4.1 Fast Non ReasoningSpaceXAI

65.15%

Gemini 2.5 Flash Lite Preview 09 2025Google

64.14%

GPT-5 NanoOpenAI

63.38%

Langston Nim Nvidia Llama 3.3 Nemotron Super 49b V1 42e84561 ThinkingTogether AI

62.37%

Magistral Medium 2509Mistral AI

62.37%

Grok 4 Fast Non ReasoningSpaceXAI

62.12%

DeepSeek V3 0324Fireworks AI

61.62%

Claude 3.5 SonnetAnthropic

59.34%

Mimo V2 FlashXiaomi

59.34%

Gemini 1.5 Pro 002Google

58.33%

100

Magistral Small 2509Mistral AI

58.33%

101

Laguna Xs.2Poolside

58.08%

102

Gemini 2.5 Flash Preview 04 17Google

57.58%

103

DeepSeek V3Fireworks AI

54.55%

104

GPT-4oOpenAI

53.79%

105

GPT-4.1 NanoOpenAI

50.76%

106

Mistral Small 2402Mistral AI

50.76%

107

Grok 2 1212SpaceXAI

50.76%

108

GPT-4oOpenAI

50.25%

109

Meta Llama Llama 3.3 70B Instruct TurboTogether AI

50.00%

110

Command A 03 2025Cohere

48.48%

111

Mistral Large 2411Mistral AI

47.73%

112

Langston Nim Nvidia Llama 3.3 Nemotron Super 49b V1 42e84561Together AI

47.73%

113

Meta Llama Llama 4 Scout 17B 16E InstructTogether AI

46.97%

114

Gemini 1.5 Flash 002Google

45.96%

115

Mistral Small 2503Mistral AI

44.19%

116

GPT-4o MiniOpenAI

44.19%

117

Claude 3.5 HaikuAnthropic

37.88%

118

Jamba Large 1.6AI21 Labs

36.11%

119

Mistral Medium 3.5Mistral AI

34.85%

120

Jamba Mini 1.6AI21 Labs

32.07%

121

Command R PlusCohere

31.06%

122

GPT-3.5 TurboOpenAI

30.56%

123

Laguna M.1Poolside

27.02%

The published Vals GPQA Diamond mirror snapshot places Gemini 3.1 Pro Preview first at 95.45%. The third row is 2.27 points behind. The broader top-10 range is 3.79 points, so many of the published results sit in a relatively narrow band.

123 models have been evaluated on Vals GPQA Diamond mirror. The benchmark falls in the External benchmark mirrors category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Vals GPQA Diamond mirror is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About Vals GPQA Diamond mirror

Year

2026

Tasks

GPQA Diamond task splits

Format

Accuracy score

Difficulty

Graduate science reasoning

BenchLM keeps this Vals-hosted GPQA Diamond table separate from canonical GPQA source records.

Vals GPQA Diamond Public benchmark source

BenchLM freshness & provenance

Version

Vals GPQA Diamond mirror 2026

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.

FAQ

What does Vals GPQA Diamond mirror measure?

Vals AI hosted GPQA Diamond view with few-shot and zero-shot chain-of-thought task splits.

Which model leads the published Vals GPQA Diamond mirror snapshot?

Gemini 3.1 Pro Preview currently leads the published Vals GPQA Diamond mirror snapshot with 95.45% vals gpqa diamond mirror score. BenchLM shows this benchmark for display only and does not use it in overall rankings.

How many models are evaluated on Vals GPQA Diamond mirror?

123 AI models are included in BenchLM's mirrored Vals GPQA Diamond mirror snapshot, based on the public leaderboard captured on July 17, 2026.

Last updated: July 17, 2026 · mirrored from the public benchmark leaderboard

Choose a model with this week’s evidence

Join 2,000+ readers for ranking moves, pricing changes, and the claims that still need proof.

One email each week. Unsubscribe anytime.