Qwen3 235B is a Alibaba open-weight catalog model. Latest preference Elo in this catalog is 1,440 (lmarena (Aug 17, 2026)). SWE-bench sits at 59.4% (swebench (Aug 17, 2026)). List output price is $1.82/1M. Context window is 131k tokens. Open-weight Qwen3 mixture-of-experts. Numbers below are dated snapshots, not a guarantee on your traffic mix.
Compare Qwen3 235B against
Currently comparing Qwen3 235B. Alternative family tiers available.
Plotted against all active catalog models (50th percentile = catalog median).
Dated snapshot metrics aggregated from official evaluators and API providers.
| Benchmark Metric | Reported Score | Observed Source |
|---|---|---|
| Preference Elo | 1,440 | lmarena · Aug 17, 2026 |
| LiveBench | 63.5% | livebench · Aug 17, 2026 |
| SWE-bench | 59.4% | swebench · Aug 17, 2026 |
| GPQA Diamond | 76.8% | seed-bootstrap · Aug 1, 2026 |
| Time to first token | 320 ms | seed-bootstrap · Aug 1, 2026 |
| Output speed | 74 tok/s | seed-bootstrap · Aug 1, 2026 |
| Input price | $0.455/1M | openrouter · Aug 17, 2026 |
| Output price | $1.82/1M | openrouter · Aug 17, 2026 |
| Context window | 131k | openrouter · Aug 17, 2026 |
Search or pick any catalog row. Suggested matchups first, then the full list.
Anthropic current default flagship. Official API $5/$25 per 1M tokens and a 1M context window (Anthropic, Jul 24 2026).
Anthropic top-tier X-High reasoning model engineered for heavy multi-step tasks. Official API $10/$50 per 1M tokens (Claude Platform pricing, Aug 2026).
OpenAI 5.6 flagship tier. Official API $5/$30 per 1M tokens (OpenAI pricing, Jul 30 2026 update left Sol unchanged).
Prior Opus generation still billed at $5/$25. Kept as a compare baseline against Opus 5.
xAI Aug 12 2026 post-training refresh of Grok 4.5. Same $2/$6 API price, 500k context, stronger agentic traces.
Follow-on Opus release. Slightly behind 4.5 on official SWE-bench bash-only in the last published sweep; stronger multi-step autonomous agent reasoning.
Current Google Pro-class multimodal model. Long context, strong coding, billed like the 3.x Pro tier.
Anthropic frontier coding and computer-use model. SWE-bench leader on the official mini-SWE-agent harness in the Feb 2026 refresh.
Auto-discovered from OpenRouter (openai/o3-mini). Preview until a second source matches.
OpenAI flagship reasoning model for 2025–26. Strong general preference Elo and multimodal coverage.
Z.ai Aug 14 2026 post-train of the GLM-5.2 744B base. Coding-plan live; open weights promised after a two-week safety review (z.ai/blog/glm-5.3).
Anthropic workhorse. Official $2/$10 per 1M tokens made permanent on Aug 10 2026.
Google frontier multimodal model with a multi-million-token context window.
OpenAI 5.6 mid tier. Official API $2/$12 per 1M after the Jul 30 2026 price cut.
OpenAI largest dense non-reasoning model with expansive world knowledge and reduced hallucinations.
Open-weight-adjacent DeepSeek flagship. High reasoning density per dollar.
Google Aug 13 2026 workhorse. Official intro price $0.75/$3.75 per 1M through Dec 31 2026; 1,048,576-token context (Google blog).
Moonshot AI frontier flagship reasoning and agent swarm model with 256k context and top-tier SWE-bench coding capability.
Workhorse Anthropic model: most of Opus coding quality at a mid-tier price.
Alibaba flagship. Strong math and multilingual code.
Auto-discovered from OpenRouter (moonshotai/kimi-k2.5). Preview until a second source matches.
Auto-discovered from OpenRouter (openai/o1). Preview until a second source matches.
Google Jul 21 workhorse. Now shares the 3.7 Flash introductory $0.75/$3.75 rate through Dec 31 2026.
Fast Google frontier-adjacent model. Near-top official SWE-bench at a fraction of Opus price.
Alibaba specialized open reasoning model competing with frontier closed reasoning models.
Zhipu flagship. Strong Chinese/English coding and agents.
Google experimental reasoning model that visualizes thoughts in real-time.
DeepSeek Jul 31 2026 price-performance SKU. Public listings put it near $0.14/$0.28 per 1M tokens.
First Claude 4 Opus generation. Baseline for opus 4 vs opus 5.
MiniMax coding model. Tied near the top of official SWE-bench bash-only in Feb 2026.
01.AI ultra-fast reasoning model delivering top LiveBench efficiency.
Auto-discovered from OpenRouter (z-ai/glm-5v-turbo). Preview until a second source matches.
ByteDance Seed 2.1 Turbo, listed on public model timelines as an Aug 10 2026 API drop.
ByteDance flagship enterprise model with ultra-low token cost and 128k context.
Meta natively multimodal open-weight flagship.
OpenAI 5.6 fast/cheap tier. Official API $0.20/$1.20 per 1M after the Jul 30 80% Luna cut.
Auto-discovered from OpenRouter (moonshotai/kimi-k2). Preview until a second source matches.
Mistral European flagship with strong function calling.
First Sonnet 4 generation. Bridge between 3.5/3.7 and Sonnet 5.
Baidu current multimodal enterprise foundation model with broad Chinese knowledge.
Alibaba balanced flagship API model with high-throughput general reasoning.
OpenAI high-speed, cost-effective reasoning model optimized for STEM, math, and code generation.
Anthropic cheap/fast Claude SKU. Official Claude Platform list price $1/$5 per 1M tokens (Anthropic Haiku page).
01.AI full-scale dense model for complex instruction following.
Alibaba dedicated open-weight code generation model with near-frontier SWE-bench Verified coding capability.
Auto-discovered from OpenRouter (qwen/qwen-2.5-coder-32b-instruct). Preview until a second source matches.
Moonshot ultra-long context model supporting up to 2 million tokens per request.
Meta open-weight Llama 4 long-context sibling of Maverick. Common public API lists sit near $0.08–$0.30 / $0.30–$0.70 per 1M; we store a conservative hosted list until OpenRouter overwrites.
Cost-efficient GPT-5 distill for high-volume agents.
Previous xAI generation before Grok 4. Kept for grok 3 vs grok 4.
Alibaba open-weight 27B drop dated Aug 14 2026. Dense enough to self-host; not a frontier MoE.
Meta flagship open-weight 405B dense foundation model with 128k context window.
DeepSeek open-weight Mixture-of-Experts coding model supporting 338 programming languages and 128k context.
Hybrid-reasoning Sonnet from 2025. Kept for historical compare pages.
ByteDance high-speed lightweight model priced at sub-cent levels.
Open-weights reasoning model trained with large-scale RL.
Previous Google long-context flagship.
Cohere enterprise RAG and tool-use model.
Previous OpenAI flagship. Still a common compare baseline on legacy pages.
Mistral code-specialist model.
Prior DeepSeek flagship. Baseline for v3 vs v4.
Previous Google speed workhorse.
Previous Meta 70B open-weight workhorse.
2024 workhorse. Still a high-intent compare against GPT-4o.
Legacy small OpenAI model. Useful as a cheap baseline.
First million-token Gemini Pro. Baseline for 1.5 vs 2.5 vs 3.x Pro.
GPT-4 Turbo 128k. Historical flagship for gpt-4 turbo vs gpt-4o / gpt-5.
Original Claude 3 flagship. Kept so opus 3 vs later Opus and vs GPT-4o still resolve.
Llama 3.1 70B instruct. Predecessor to 3.3 70B and Llama 4.
Previous cheap Claude. Haiku 4.5 is the current $1/$5 SKU.
OpenAI flagship diffusion-transformer image synthesis model with supreme prompt adherence, realistic textures, and complex text composition.
Fast distilled tier of GPT Image 2 designed for high-throughput interactive creative workflows at 70% lower price.
Previous OpenAI image generation flagship. High fidelity with proven enterprise reliability.
Legacy OpenAI image model benchmark baseline. Retained for historical comparisons.
State-of-the-art cinematic image synthesis foundation model known for photorealistic lighting and aesthetic composition.
Google DeepMind frontier multimodal image generation flagship with Deep Research and complex multi-object spatial reasoning.
Google high-speed multimodal generative model delivering top Elo performance at half the latency and cost of Pro.
Sub-1.5 second ultra-low latency tier for real-time applications and mobile game asset pipelines.
Black Forest Labs maximum-capacity flow-matching model with photoreal anatomy and leading typography fidelity.
Flexible step-distilled variant of FLUX.2 offering 99% of Max quality with 45% faster generation speeds.
First-generation professional API model from Black Forest Labs, renowned for 6x faster generation than original FLUX.1 Pro.
Open-weights non-commercial base model with 12B parameters, serving as the foundation for the open-source fine-tuning ecosystem.
Apache 2.0 4-step distilled open weights model optimized for local inference and instant preview generation.
Industry-leading typography, graphic design, and in-image text layout model with precise kerning and complex banner generation.
Open-weight release of Ideogram 4.0, bringing world-class text rendering to self-hosted enterprise infrastructure.
Specialized generative model for vector art, icon sets, illustrations, and commercial design assets with native brand color matching.
High-speed utility tier of Recraft V4.1 tailored for rapid asset production at $0.035/image.
High-resolution real-time generation model tuned for artistic composition and dynamic concept art.
Sub-2 second interactive generation engine delivering ultra-affordable $15/1k image generation.
Alibaba multimodal image synthesis foundation model with strong bilingual Chinese/English typography and cultural asset fidelity.
Alibaba high-efficiency open video/image unified transformer backbone.
ByteDance flagship image generation model with high photorealism and fine facial structure rendering.
High-value ByteDance image foundation model at $30/1k images.
Midjourney v7 premier creative image generation model with unparalleled stylistic nuances, coherent hands/anatomy, and cinematic grading.
Previous standard in aesthetic digital art generation, retained as a historical compare baseline.
Microsoft AI proprietary foundation image model for Copilot Studio and enterprise creative workflows.
Cost-optimized Microsoft AI image model delivering sub-2 second responses at $20/1k images.
HiDream high-definition visual generator optimized for photorealism and accurate complex lighting.
Luma Labs universal 3D-aware image synthesis engine with spatial geometry consistency.
Plain-English methodology and leaderboard answers
Elo is a crowd vote on which hidden answer people liked more — not a school test. What is Elo?