Live Benchmark Intelligence|Q3 2026
AI Model Leaderboard &
Empirical Comparisons
Independent, verifiable rankings of GPT, Claude, Gemini, Llama, DeepSeek, and 45+ multi-modality AI models. Evaluated continuously by Composite TrueSkill, coding benchmarks, output speed, and live API pricing.
45+ Evaluated Models6 AI Modalities100% Verified Specs
LIVE BENCHMARK EVALQ3 2026
Leads reasoning
GPT-5.6 Sol
94.6% GPQA
Wins at coding
Claude Opus 5.1
2,710 Arena ELO
Cheapest Top 10
DeepSeek-V4 Pro
$0.48 / M tok
Fastest agentic output
Gemini 3.8 Flash
348-620 tok/s
Longest context
Grok 4.6
2.0M Tokens
Best open-weights
Kimi K3
93.5% GPQA
Best AI for Coding
Code generation, debugging & refactoring
Best AI for Writing
Prose, tone matching & creative content
Best AI for Research
Analysis, synthesis & deep logical tasks
Best Image Generation
Text-to-image diffusion & editing
Efficiency Index
Price vs Performance (Pareto Frontier)
Blended token cost (8:1 input/output ratio) plotted against TrueSkill score. Models on the green line are Pareto-efficient.
PARETO EFFICIENT (HIGH ROI)
![FLUX.1 [pro] logo](https://img.logo.dev/blackforestlabs.ai?token=pk_LLpejae2QeOis7Keq3ntaw&size=64){kind=logo}
$0.20 / 1M$0.50$1.00$2.00$5.00$10.00$15.00+ / 1M tokens
Index Ranking
Model Performance vs Blended Price
01
ChatGPT Astra (GPT-6 Astra)OpenAI
61.2
$20.00
02
Gemini 3.8 FlashGoogle
59
$1.12
03
Claude Opus 5.1Anthropic
58.8
$15.00
04
Claude Fable 5.1Anthropic
58.2
$12.00
05
Muse Spark 1.3Meta
54.8
$1.58
06
GPT-5.6 SolOpenAI
57.4
$7.78
07
Claude Opus 5Anthropic
56.2
$7.22
08
Claude Fable 5Anthropic
56.1
$14.40
09
Claude Mythos PreviewAnthropic
56
Free
10
Kimi K3Moonshot AI
54.9
$4.33
Master Matrix
AI Model Comparison Table
| Rank | Model & Provider | Score | Reasoning | Coding | Arena ELO | Architecture | Context | Speed | Pricing $/1M | Action |
|---|---|---|---|---|---|---|---|---|---|---|
| 1NEW | ChatGPT Astra (GPT-6 Astra) OpenAI | 61.2 | 63.5 | 57.8 | 2,695 | Recurrent Depth (~1.8T Equiv) | 1.2M | 95 tok/s | $20.00 | Compare |
| 1NEW | Gemini 3.8 Flash Google | 59 | 56.5 | 54.2 | 2,190 | Sparse MoE (~140B Active) | 1.0M | 348 tok/s | $1.12 | Compare |
| 2NEW | Claude Opus 5.1 Anthropic | 58.8 | 57.2 | 51.5 | 2,710 | Dense (~250B) | 1.0M | 64 tok/s | $15.00 | Compare |
| 3NEW | Claude Fable 5.1 Anthropic | 58.2 | 56 | 53.5 | 2,160 | MoE (~380B) | 1.0M | 86 tok/s | $12.00 | Compare |
| 1 | GPT-5.6 Sol OpenAI | 57.4 | 56.8 | 50.6 | 2,134 | MoE (~1.8T) | 1.1M | 102 tok/s | $7.78 | Compare |
| 2 | Claude Opus 5 Anthropic | 56.2 | 55.3 | 42.7 | 2,668 | Dense (~220B) | 1.0M | 58 tok/s | $7.22 | Compare |
| 3 | Claude Fable 5 Anthropic | 56.1 | 53.5 | 48.8 | 2,003 | MoE (~350B) | 1.0M | 78 tok/s | $14.40 | Compare |
| 4NEW | Claude Mythos Preview Anthropic | 56 | 56.8 | 46.6 | 2,190 | Research Tier | 1.0M | 45 tok/s | Free / Open | Compare |
| 5 | Kimi K3 Moonshot AI | 54.9 | 53.9 | 45.9 | 1,816 | 2.8T (MoE) | 1.0M | 135 tok/s | $4.33 | Compare |
| 8NEW | Muse Spark 1.3 Meta | 54.8 | 54.5 | 47.2 | 1,840 | MoE (~110B Active) | 1.0M | 245 tok/s | $1.58 | Compare |
| 6NEW | GLM-5.3 Zhipu AI | 54.7 | 54.9 | 45.4 | 1,780 | 753B (MoE) | 1.0M | 142 tok/s | $1.73 | Compare |
| 7NEW | DeepSeek-V4-Pro-0813 DeepSeek | 54.5 | 52 | 44.3 | 1,980 | 1.6T (MoE) | 1.0M | 199 tok/s | $0.48 | Compare |
| 8NEW | Qwen3.8 Max Alibaba Cloud / Qwen Team | 53.2 | 52 | 42.1 | 1,850 | 2.4T (MoE) | 1.0M | 160 tok/s | $0.85 | Compare |
| 9 | GPT-5.6 Terra OpenAI | 52.8 | 51.1 | 46.4 | 1,185 | MoE (~500B) | 1.1M | 119 tok/s | $3.11 | Compare |
| 10 | Claude Opus 4.8 Anthropic | 51.9 | 51.4 | 43.8 | 1,689 | Dense | 1.0M | 125 tok/s | $7.22 | Compare |
| 11 | Muse Spark 1.1 Meta | 51.7 | 52.3 | 37.8 | 1,070 | Dense (~70B) | 1.0M | 229 tok/s | $1.58 | Compare |
| 12NEW | Gemini 3.7 Flash Google | 51.1 | 49.9 | 38.6 | 1,720 | Sparse MoE | 1.0M | 621 tok/s | $1.08 | Compare |
| 13 | Claude Sonnet 5 Anthropic | 49.6 | 49 | 40 | 1,588 | Dense (~100B) | 1.0M | 59 tok/s | $2.89 | Compare |
| 14 | GPT-5.5 OpenAI | 49.3 | 48.6 | 41.1 | 2,124 | MoE | 1.1M | 49 tok/s | $7.78 | Compare |
| 15NEW | DeepSeek-V4-Flash-Vision-Exp DeepSeek | 48.7 | 45.3 | 38.5 | 1,650 | Vision MoE | 1.0M | 310 tok/s | $0.27 | Compare |
| 16NEW | Grok 4.5 xAI | 48.2 | 47.8 | 37.2 | 1,540 | MoE (~300B) | 2.0M | 110 tok/s | $2.00 | Compare |
Direct Head-to-Head
Popular 1v1 Model Comparisons
Direct head-to-head empirical evaluations across speed, pricing, and verified benchmark scores.
56 Total MatchupsLeaderboard
Frontier Autonomous Agent & Engineering ClashVerified
VS
ChatGPT Astra vs Claude Fable 5.1
$20.00/M
Blended Price$12.00/M
View Full Comparison
Autonomous Agent Monarch vs Apex Human PickVerified
VS
ChatGPT Astra vs Claude Opus 5.1
$20.00/M
Blended Price$15.00/M
View Full Comparison
Agentic Coding vs Flagship Cognitive LogicVerified
VS
Gemini 3.8 Flash vs GPT-5.6 Sol
$1.12/M
Blended Price$7.78/M
View Full Comparison
Open Weights Coding vs Cognitive ApexVerified
VS
Muse Spark 1.3 vs GPT-5.6 Sol
$1.58/M
Blended Price$7.78/M
View Full Comparison
Autonomous Software Engineering FaceoffVerified
VS
Gemini 3.8 Flash vs Claude Fable 5.1
$1.12/M
Blended Price$12.00/M
View Full Comparison
Open-Weights Coding & Speed SuperclashVerified
VS
Muse Spark 1.3 vs DeepSeek-V4 Pro
$1.58/M
Blended Price$0.48/M
View Full Comparison
Showing 1 to 6 of 56 comparisons
Taxonomy Explorer
Browse AI Models by Category
By Capability
By Modality
Intelligence Feed
Latest AI Model Research & Releases
Frontier Release
OpenAI Releases GPT-6 Astra: 99.9% ARC-AGI-3 & Critical Cybersecurity Rating
OpenAI deploys GPT-6 Astra featuring recurrent depth architecture, Symphony cross-modal encoding, 72.6% OSWorld 2.0 autonomous computer control, and 98.0% FrontierMath Tier 4.
Softzar AI LabRead Teardown
Frontier Release
Gemini 3.8 Flash Released: 90.8% Terminal-Bench & 71.0% DeepSWE at $0.75/M
Google officially deploys Gemini 3.8 Flash featuring 90.8% Terminal-Bench 2.1 shell coding, 71.0% DeepSWE v1.1, and 348-620 tokens/sec generation on TPU v6e.
Softzar AI LabRead Teardown
Agent Benchmarks
Claude Fable 5.1 & Mythos 5.1: Non-Shortcut Long-Horizon Agent Engineering
Anthropic launches Fable 5.1 with deep agentic multi-file repository refactoring, reaching 62.8% on SWE-Bench Pro with strict anti-shortcut evaluation.
Softzar AI LabRead Teardown
Open Weights
Meta Launches Muse Spark 1.3: 245 tok/s Throughput & Dual Engine Architecture
Meta releases Muse Spark 1.3 open-weights flagship with dual Max and XHigh engines, 1.0M context, and deep Muse Code integration at $1.58/M blended.
Softzar AI LabRead Teardown
Release Analysis
DeepSeek-V4-Flash-Vision-Exp: Vision at Flash Price
DeepSeek's first official V4-Flash vision API. Same $0.22/$0.66 off-peak list as V4-Flash, 1M context. Self-reported Chartography 64.3 and TerminalBench 83.9.
Softzar AI LabRead Teardown