Compare · ModelsLive · 3 picked · head to head
Claude Opus 5 vs Gemini 3.8 Flash vs GPT-6 Astra
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
GPT-6 Astra wins on 22/29 benchmarks
GPT-6 Astra wins 22 of 29 shared benchmarks. Leads in knowledge · coding · math.
Category leads
agentic·Claude Opus 5knowledge·GPT-6 Astracoding·GPT-6 Astrageneral·Claude Opus 5math·GPT-6 Astraspeed·GPT-6 Astrareasoning·GPT-6 Astra
Hype vs Reality
Attention vs performance
Claude Opus 5
#21 by perf·#9 by attention
Gemini 3.8 Flash
#63 by perf·#5 by attention
GPT-6 Astra
#9 by perf·#1 by attention
Best value
Gemini 3.8 Flash
5.5x better value than Claude Opus 5
Claude Opus 5
4.6 pts/$
$15.00/M
Gemini 3.8 Flash
25.5 pts/$
$2.25/M
GPT-6 Astra
2.6 pts/$
$30.00/M
Vendor risk
Who is behind the model
Anthropic
$965.0B·Tier 1
Google DeepMind
$4.20T·Tier 1
OpenAI
$840.0B·Tier 1
Head to head
29 benchmarks · 3 models
Claude Opus 5Gemini 3.8 FlashGPT-6 Astra
APEX-Agents
Claude Opus 5 leads by +1.1
APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments.
Claude Opus 5
65.8
Gemini 3.8 Flash
64.3
GPT-6 Astra
64.7
Chess Puzzles
GPT-6 Astra leads by +11.6
Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities.
Claude Opus 5
39.0
Gemini 3.8 Flash
59.0
GPT-6 Astra
70.5
Deepswe
GPT-6 Astra leads by +0.3
Claude Opus 5
73.7
Gemini 3.8 Flash
73.8
GPT-6 Astra
74.1
Dtbench
Claude Opus 5 leads by +0.9
Claude Opus 5
96.5
Gemini 3.8 Flash
92.9
GPT-6 Astra
95.5
Frontiercode
Claude Opus 5 leads by +0.1
Claude Opus 5
53.4
Gemini 3.8 Flash
41.2
GPT-6 Astra
53.3
FrontierMath-Tier-4-v2-Private
GPT-6 Astra leads by +24.4
Claude Opus 5
73.2
Gemini 3.8 Flash
21.9
GPT-6 Astra
97.6
FrontierMath-Tiers-1-3-v2-Private
GPT-6 Astra leads by +8.1
Claude Opus 5
85.6
Gemini 3.8 Flash
68.4
GPT-6 Astra
93.7
Frontierswe
GPT-6 Astra leads by +13.5
Claude Opus 5
52.0
Gemini 3.8 Flash
19.6
GPT-6 Astra
65.5
Furniture Assembly
GPT-6 Astra leads by +27.4
Claude Opus 5
44.0
Gemini 3.8 Flash
2.4
GPT-6 Astra
71.4
GPQA diamond
GPT-6 Astra leads by +0.5
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Opus 5
91.8
Gemini 3.8 Flash
93.9
GPT-6 Astra
94.4
Lmca
Claude Opus 5 leads by +0.1
Claude Opus 5
75.8
Gemini 3.8 Flash
62.2
GPT-6 Astra
75.7
Mystery Game Puzzles
GPT-6 Astra leads by +27.5
Claude Opus 5
54.8
Gemini 3.8 Flash
41.6
GPT-6 Astra
82.4
OTIS Mock AIME 2024-2025
GPT-6 Astra leads by +1.1
OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills.
Claude Opus 5
98.9
Gemini 3.8 Flash
98.9
GPT-6 Astra
100.0
Proofbench
Claude Opus 5
99.0
Gemini 3.8 Flash
48.0
GPT-6 Astra
99.0
SimpleQA Verified
GPT-6 Astra leads by +5.9
SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information.
Claude Opus 5
59.9
Gemini 3.8 Flash
69.7
GPT-6 Astra
75.6
Artificial Analysis · CritPt
GPT-6 Astra leads by +13.4
Gemini 3.8 Flash
18.3
GPT-6 Astra
31.7
Artificial Analysis · GDPval
GPT-6 Astra leads by +6.5
Gemini 3.8 Flash
45.6
GPT-6 Astra
52.1
Artificial Analysis · GPQA Diamond
GPT-6 Astra leads by +0.8
Gemini 3.8 Flash
95.3
GPT-6 Astra
96.1
Artificial Analysis · Humanity's Last Exam
GPT-6 Astra leads by +6.9
Gemini 3.8 Flash
47.8
GPT-6 Astra
54.7
Artificial Analysis · Long Context Reasoning
Gemini 3.8 Flash leads by +0.6
Gemini 3.8 Flash
81.3
GPT-6 Astra
80.7
Artificial Analysis · MMMU Pro
GPT-6 Astra leads by +1.3
Gemini 3.8 Flash
85.6
GPT-6 Astra
86.9
Artificial Analysis · Quality Index
GPT-6 Astra leads by +11.7
Gemini 3.8 Flash
40.9
GPT-6 Astra
52.7
Artificial Analysis · SciCode
Gemini 3.8 Flash leads by +0.1
Gemini 3.8 Flash
56.6
GPT-6 Astra
56.5
ARC-AGI
GPT-6 Astra leads by +1.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Claude Opus 5
97.5
GPT-6 Astra
98.5
ARC-AGI-2
GPT-6 Astra leads by +4.6
ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data.
Claude Opus 5
90.4
GPT-6 Astra
95.0
Balrog
GPT-6 Astra leads by +4.9
Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning.
Claude Opus 5
63.4
GPT-6 Astra
68.3
Ebr Bench
GPT-6 Astra leads by +30.5
Claude Opus 5
45.7
GPT-6 Astra
76.2
HLE
GPT-6 Astra leads by +10.8
HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%.
Gemini 3.8 Flash
41.7
GPT-6 Astra
52.5
WeirdML
GPT-6 Astra leads by +1.8
WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns.
Claude Opus 5
91.8
GPT-6 Astra
93.6
Full benchmark table
| Benchmark | Claude Opus 5 | Gemini 3.8 Flash | GPT-6 Astra |
|---|---|---|---|
APEX-Agents APEX-Agents · evaluates AI agents on complex, multi-step tasks requiring planning, tool use, and autonomous decision-making in realistic environments. | 65.8 | 64.3 | 64.7 |
Chess Puzzles Chess Puzzles · tests strategic and tactical reasoning by having models solve chess puzzle positions, evaluating lookahead and pattern recognition abilities. | 39.0 | 59.0 | 70.5 |
Deepswe | 73.7 | 73.8 | 74.1 |
Dtbench | 96.5 | 92.9 | 95.5 |
Frontiercode | 53.4 | 41.2 | 53.3 |
FrontierMath-Tier-4-v2-Private | 73.2 | 21.9 | 97.6 |
FrontierMath-Tiers-1-3-v2-Private | 85.6 | 68.4 | 93.7 |
Frontierswe | 52.0 | 19.6 | 65.5 |
Furniture Assembly | 44.0 | 2.4 | 71.4 |
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 91.8 | 93.9 | 94.4 |
Lmca | 75.8 | 62.2 | 75.7 |
Mystery Game Puzzles | 54.8 | 41.6 | 82.4 |
OTIS Mock AIME 2024-2025 OTIS Mock AIME 2024-2025 · simulated American Invitational Mathematics Examination problems testing advanced problem-solving skills. | 98.9 | 98.9 | 100.0 |
Proofbench | 99.0 | 48.0 | 99.0 |
SimpleQA Verified SimpleQA Verified · short factual questions with verified answers, measuring factual accuracy and the tendency to hallucinate or provide incorrect information. | 59.9 | 69.7 | 75.6 |
Artificial Analysis · CritPt | — | 18.3 | 31.7 |
Artificial Analysis · GDPval | — | 45.6 | 52.1 |
Artificial Analysis · GPQA Diamond | — | 95.3 | 96.1 |
Artificial Analysis · Humanity's Last Exam | — | 47.8 | 54.7 |
Artificial Analysis · Long Context Reasoning | — | 81.3 | 80.7 |
Artificial Analysis · MMMU Pro | — | 85.6 | 86.9 |
Artificial Analysis · Quality Index | — | 40.9 | 52.7 |
Artificial Analysis · SciCode | — | 56.6 | 56.5 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | 97.5 | — | 98.5 |
ARC-AGI-2 ARC-AGI-2 · the second iteration of the Abstraction and Reasoning Corpus, testing novel pattern recognition and abstract reasoning without prior training data. | 90.4 | — | 95.0 |
Balrog Balrog · benchmarks AI agents on text-based adventure games, testing language understanding, strategic planning, and long-horizon reasoning. | 63.4 | — | 68.3 |
Ebr Bench | 45.7 | — | 76.2 |
HLE HLE (Humanity's Last Exam) · a reasoning benchmark designed to be the hardest public evaluation of AI. Questions span mathematics, physics, philosophy, and logic · curated to be at or beyond the frontier of human expert capability. Tested with and without tool augmentation. Claude Opus 4.7 scores 46.9% without tools and 54.7% with tools · making it one of the few benchmarks where the top score is below 60%. | — | 41.7 | 52.5 |
WeirdML WeirdML · tests models on unusual and adversarial machine learning tasks that require creative problem-solving beyond standard patterns. | 91.8 | — | 93.6 |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| $5.00 | $25.00 | 1.0M tokens (~500 books) | $100.00 | |
| $0.75 | $3.75 | 1.0M tokens (~524 books) | $15.00 | |
| $10.00 | $50.00 | 1.1M tokens (~525 books) | $200.00 |