Compare · ModelsLive · 3 picked · head to head
Claude Mythos Preview vs Gemini 3.1 Pro Preview vs GPT-5.5
Side by side · benchmarks, pricing, and signals you can act on.
Winner summary
Claude Mythos Preview wins on 4/6 benchmarks
Claude Mythos Preview wins 4 of 6 shared benchmarks. Leads in knowledge · agentic.
Category leads
knowledge·Claude Mythos Previewcoding·GPT-5.5reasoning·Gemini 3.1 Pro Previewagentic·Claude Mythos Preview
Hype vs Reality
Attention vs performance
Claude Mythos Preview
#5 by perf·no signal
Gemini 3.1 Pro Preview
#137 by perf·#6 by attention
GPT-5.5
#2 by perf·#4 by attention
Best value
Gemini 3.1 Pro Preview
1.4x better value than GPT-5.5
Claude Mythos Preview
n/a
no price
Gemini 3.1 Pro Preview
6.9 pts/$
$7.00/M
GPT-5.5
4.9 pts/$
$17.50/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
6 benchmarks · 3 models
Claude Mythos PreviewGemini 3.1 Pro PreviewGPT-5.5
GPQA diamond
Claude Mythos Preview leads by +0.9
Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs.
Claude Mythos Preview
94.5
Gemini 3.1 Pro Preview
92.6
GPT-5.5
93.6
Terminal Bench
GPT-5.5 leads by +0.7
Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence.
Claude Mythos Preview
82.0
Gemini 3.1 Pro Preview
80.2
GPT-5.5
82.7
ARC-AGI
Gemini 3.1 Pro Preview leads by +3.0
ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization.
Gemini 3.1 Pro Preview
98.0
GPT-5.5
95.0
HLE
Claude Mythos Preview leads by +13.1
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%.
Claude Mythos Preview
56.8
Gemini 3.1 Pro Preview
43.7
OSWorld
Claude Mythos Preview leads by +0.9
OSWorld · tests AI agents on real-world computer tasks across operating systems, including web browsing, file management, and application use.
Claude Mythos Preview
79.6
GPT-5.5
78.7
SWE-Bench verified
Claude Mythos Preview leads by +18.3
SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability.
Claude Mythos Preview
93.9
Gemini 3.1 Pro Preview
75.6
Full benchmark table
| Benchmark | Claude Mythos Preview | Gemini 3.1 Pro Preview | GPT-5.5 |
|---|---|---|---|
GPQA diamond Graduate-Level Google-Proof QA (Diamond set) · expert-crafted questions in physics, biology, and chemistry that are difficult even for domain PhDs. | 94.5 | 92.6 | 93.6 |
Terminal Bench Terminal-Bench 2.0 · evaluates AI agents on real terminal-based coding tasks · writing scripts, debugging, running tests, and managing projects entirely through command-line interaction. Tests both code quality and terminal fluency. Claude Opus 4.7 scores 69.4%, demonstrating significant agentic terminal competence. | 82.0 | 80.2 | 82.7 |
ARC-AGI ARC-AGI · the original Abstraction and Reasoning Corpus, testing whether AI can solve novel visual pattern recognition tasks without memorization. | — | 98.0 | 95.0 |
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%. | 56.8 | 43.7 | — |
OSWorld OSWorld · tests AI agents on real-world computer tasks across operating systems, including web browsing, file management, and application use. | 79.6 | — | 78.7 |
SWE-Bench verified SWE-bench Verified · 500 human-validated tasks from 12 real Python repositories (Django, Flask, scikit-learn, sympy, and others). Each task requires the model to produce a git patch that resolves a real GitHub issue and passes the test suite. The verified subset eliminates ambiguous tasks from the original SWE-bench. Claude Mythos Preview leads at 93.9%, crossing 90% for the first time in 2026. Opus 4.6 scores 80.8%. The benchmark remains the most-cited evaluation for code-generation capability. | 93.9 | 75.6 | — |
Pricing · per 1M tokens · projected $/mo at 10M tokens
| Model | Input | Output | Context | Projected $/mo |
|---|---|---|---|---|
| — | — | 1.0M tokens (~500 books) | — | |
| $2.00 | $12.00 | 1.0M tokens (~524 books) | $45.00 | |
| $5.00 | $30.00 | 400K tokens (~200 books) | $112.50 |