GPQA
gpqaA challenging dataset of 448 multiple-choice questions written by domain experts in biology, physics, and chemistry. Questions are Google-proof and extremely difficult, with PhD experts reaching 65% accuracy.
VENICE ATLAS · 2026 EDITION · MODELS & EVIDENCE
AI model guide
Find the right model
for your task
Explore models from Venice AI and compare their pricing, capabilities, context windows, and benchmark results side by side.
Start with your use case, not a winner.
Use this guide to narrow your options. Text models and their evidence are shown here; image, video, audio, and other endpoints are available in the full catalog and Showcase.
View featured models ↓Explore Venice AI
140text models available
764benchmarks available
94models with comparable evidence
Featured models
A quick view of the
leading options
Compare overall position, category ratings, context window, and native pricing for models available through Venice AI.
Each model is rated on programming, reasoning, mathematics, vision, multimodal tasks, and agents. Rankings are updated with every data refresh.
Top 10 · ordered by overall position
| Position | Model | Score | Reasoning | Coding | Agents | Context |
|---|---|---|---|---|---|---|
#1→ | Claude Opus 5.5 | 60.33 | 59.33 | 49.72 | 45.42 | 1M |
#2↑ 1 | Claude Sonnet 5.5 | 58.30 | 57.07 | 47.46 | 43.21 | 1M |
#3↓ 1 | GPT-6 Astra | 58.07 | 57.13 | 46.23 | 44.93 | 1.1M |
#4NEW | Gemini 4 Argon | 54.98 | 53.03 | 42.36 | 41.06 | 1M |
#5↓ 1 | Claude Opus 5 | 53.74 | 53.96 | 38.20 | 38.33 | 1M |
#6↓ 1 | GPT-6.1 Sol | 53.51 | 51.54 | 38.40 | 37.87 | 1.1M |
#7↓ 1 | Claude Fable 5.1 | 53.22 | 50.74 | 34.18 | 38.21 | 1M |
#8↓ 2 | GPT-5.6 Sol | 53.20 | 52.61 | 42.38 | 39.12 | 1M |
#9↓ 1 | Claude Fable 5 | 53.08 | 51.12 | 42.66 | 38.66 | 1M |
#10↓ 1 | Kimi K3 | 51.83 | 51.14 | 39.85 | 37.25 | 1M |
How to read this listThe overall score is published by LLM-Stats.Category scores are shown only when there's benchmark data.Arrows show how each rank changed since the last update.
Top 10 · private models only · prompts never leave Venice infrastructure
| Position | Model | Score | Reasoning | Coding | Agents | Context |
|---|---|---|---|---|---|---|
#1→ | Kimi K3 | 51.83 | 51.14 | 39.85 | 37.25 | 1M |
#2NEW | DeepSeek v4.1 Flash (1017) | 51.04 | 48.62 | 39.13 | 38.56 | 1M |
#3↓ 1 | GLM 5.3 | 50.77 | 50.49 | 39.22 | 36.88 | 1M |
#4↓ 1 | DeepSeek V4 Pro 0813 | 50.26 | 48.54 | 38.15 | 35.25 | 1M |
#5↓ 1 | DeepSeek V4.1 Flash | 49.77 | 47.73 | 40.30 | 38.16 | 1M |
#6↓ 1 | Hy4 Preview | 49.57 | 49.69 | 36.99 | 34.80 | 1.0M |
#7↓ 1 | GLM 5.3 Flash | 48.64 | 47.78 | 59.90 | 34.66 | 1.0M |
#8↓ 1 | Grok 4.7 | 46.57 | 63.20 | 34.84 | 33.79 | 500K |
#9NEW | Grok 4.5 | 45.31 | 44.46 | 49.50 | 30.29 | 500K |
#10NEW | Muse Spark Open | 44.93 | 45.23 | 32.96 | 30.56 | 1.0M |
Why private?Private models run entirely on Venice infrastructure — the model provider never sees your prompt content.Scores come from the same benchmark evidence as the overall leaderboard.
Use the evidence carefully
Good signals,
but not the full story
Benchmarks measure specific tasks under specific conditions. A model leading one benchmark is not automatically the best choice for every application.
What the scores show
Benchmark scores show performance on a defined task using a defined evaluation setup. Rankings preserve the categories and evidence published by LLM-Stats.
What the scores don't show
Your decision also depends on your workload, prompts, latency, price, reliability, context needs, and the model's behavior in your application.
How to use this guide
Use the catalog for availability and price, rankings for task-specific signals, the benchmark library for methodology, and model pages for detailed provenance.
Benchmark library
Know what each test measures
Original benchmark scores and price comparisons are shown separately. Check the methodology before using a score as a recommendation.
Humanity's Last Exam
humanity's-last-examHumanity's Last Exam (HLE) is a multi-modal academic benchmark with 2,500 questions across mathematics, humanities, and natural sciences, designed to test LLM capabilities at the frontier of human knowledge with unambiguous, verifiable solutions
SWE-Bench Pro
swe-bench-proSWE-Bench Pro is an advanced version of SWE-Bench that evaluates language models on complex, real-world software engineering tasks requiring extended reasoning and multi-step problem solving.
61 modelstextCatalog source
Terminal-Bench 2.0
terminal-bench-2Terminal-Bench 2.0 is an updated benchmark for testing AI agents' tool use ability to operate a computer via terminal. It evaluates how well models can handle real-world, end-to-end tasks autonomously, including compiling code, training models, setting up servers, system administration, security tasks, data science workflows, and cybersecurity vulnerabilities.
53 modelstextCatalog source
BrowseComp
browsecompBrowseComp is a benchmark comprising 1,266 questions that challenge AI agents to persistently navigate the internet in search of hard-to-find, entangled information. The benchmark measures agents' ability to exercise persistence in information gathering, demonstrate creativity in web navigation, and find concise, verifiable answers. Despite the difficulty of the questions, BrowseComp is simple and easy-to-use, as predicted answers are short and easily verifiable against reference answers.
MMMU-Pro
mmmu-proA more robust multi-discipline multimodal understanding benchmark that enhances MMMU through a three-step process: filtering text-only answerable questions, augmenting candidate options, and introducing vision-only input settings. Achieves significantly lower model performance (16.8-26.9%) compared to original MMMU, providing more rigorous evaluation that closely mimics real-world scenarios.
MMLU-Pro
mmlu-proA more robust and challenging multi-task language understanding benchmark that extends MMLU by expanding multiple-choice options from 4 to 10, eliminating trivial questions, and focusing on reasoning-intensive tasks. Features over 12,000 curated questions across 14 domains and causes a 16-33% accuracy drop compared to original MMLU.
IFEval
ifevalInstruction-Following Evaluation (IFEval) benchmark for large language models, focusing on verifiable instructions with 25 types of instructions and around 500 prompts containing one or more verifiable constraints
Continue exploring
Get the full picture. Visit our pages for complete rankings, side-by-side comparisons, or examples of image and video outputs.