AI model comparison — text, image and video
The leading AI models from the US, Europe and China — text, image and video generation — compared by quality (market benchmarks), cost and open-source status, always separating remote (API) and local (on your hardware) models.
Data as of 2026-11-12 · automated research (Artificial Analysis, LMArena, official pricing) — verify before deciding.
📊 How quality is measured — three indices
We show quality three complementary ways. Here is how each index is built before you read the charts:
① Zendoric Quality (0-100) = equal-thirds average of SWE-bench-Pro (33% · real software development, checked against the maker) + LMArena (33% · human preference, normalised Elo) + Terminal-Bench (33% · agentic terminal capability). If a model lacks one of the three, its weight is shared among those present (at least two required).
② AA Index (Artificial Analysis Intelligence Index, 0-100) = a broader composite index (reasoning, science, code, maths). It gives a second reading: depending on how you measure, the maker ranking changes.
③ Cybersecurity (0-100) = capability on expert cyber tasks (hard «unguided pass@1» protocol: vuln-research and realistic exploitation). We use a non-saturated metric (the top is around 71, not 100, leaving headroom), not the Cybench «pass@k» the frontier already saturates. Sources: UK AISI, NIST-CAISI, CVE-Bench. We frame it as capability and risk, not an offensive ranking; where there is no direct eval it is estimated «est.».
📈 Zendoric Quality over time (frontier makers)
Quality index (0-100) of the top makers (their best model), last 24 months. Dashed line = quality estimated from the AA Index (labs without SWE-bench-Pro). Updated daily.
📈 AA Index over time (frontier makers)
AA Index (Artificial Analysis Intelligence Index, 0-100) of the top makers (their best model), last 24 months. It is a broader composite index (reasoning, science, code, maths) than ours. The historical series is reconstructed by anchoring each maker's trajectory to its current AA. Updated daily.
🛡️ Cybersecurity over time (frontier makers)
Cybersecurity index of each maker's best model, relative to the US frontier (= 100). Since October 2026 the scale is no longer UK AISI's expert CTF —saturated: GPT-5.6 Sol scores 95%— but the distance to the frontier on the SAME tests public evaluators use (UK AISI and US CAISI-NIST: ExploitBench, SEC-Bench Pro, ExploitGym, cyber range). The earlier series was rescaled to stay comparable. Confidence → high: Anthropic, OpenAI, Zhipu AI · medium: Google, Moonshot AI, xAI · low: Alibaba, DeepSeek, Meta, Mistral AI. No model-level data (not drawn): Microsoft. The whole series is estimated («est.»). We frame it as capability and RISK to govern, not an offensive ranking. Reviewed on 2026-10-15; revised when new evaluations are published.
💰 Zendoric Quality vs cost
Flagship models of the top makers by quality (a maker may have several, e.g. Anthropic: Opus 5, Opus 4.8 and Fable 5). HIGHER = more quality; LEFT = cheaper (log axis). Hollow dot = quality estimated (AA Index). Colour by maker.
💰 AA Index vs cost
Same format as the quality/cost chart, but the vertical axis is the AA Index. HIGHER = more capability; LEFT = cheaper (log axis). Hollow dot = estimated AA (Terminal-Bench/SWE-Pro). Colour by maker.
🏁 Efficient frontier — quality vs cost
Every cloud model with data. Y = AA Index; X = output cost, log and REVERSED: further RIGHT is cheaper — the top-right corner is ideal. The green line joins the efficient frontier: models with no alternative that is both better and cheaper; picking off the line is only justified by non-price factors (ecosystem, context, open source). Hollow dot = estimated AA. Hover a dot for details.
🏆 Zendoric Quality (SW dev + arena + agentic)
| Model | Zendoric Quality | SWE-bench-Pro | DeepSWE | LMArena | Terminal-Bench | LiveCodeBench | GPQA | ARC-AGI-2 |
|---|---|---|---|---|---|---|---|---|
| 🇨🇳 Kimi K3Moonshot AI · China | 90.9 | — | 68.5 | 1487 | 88.3 | 68.0 | 93.5 | — |
| 🇺🇸 Claude Mythos 5.1 (acceso restringido)Anthropic · USA | 89.3 | 80.0 | — | 1507 | 88.0 | 91.7 | 94.1 | — |
| 🆕 🇺🇸 Gemini 4 CarbonGoogle · USA | 82.6 | — | — | 1528 | 65.2 | — | — | — |
| 🆕 🇺🇸 Gemini 4 ArgonGoogle · USA | 78.7 | — | 77.9 | 1525 | 57.4 | — | 96.5 | — |
| 🇨🇳 GLM-5.3Zhipu AI · China | 76.9 | 62.1 | 69.0 | 1475 | 81.0 | 80.2 | 78 | 7 |
| 🆕 🇺🇸 GPT-6.1 SolOpenAI · USA | 76.0 | — | 75.2 | 1482 | 60.91 | — | 95.4 | — |
| 🇺🇸 Claude Opus 5.5Anthropic · USA | 75.8 | 79.2 | 73.6 | 1490 | 53.33 | 86.4 | 93.2 | — |
| 🇺🇸 Claude Sonnet 5.5Anthropic · USA | 74.7 | 63.2 | 53.8 | 1461 | 80.4 | — | 83 | 12 |
| 🆕 🇺🇸 Muse WonderMeta · USA | 74.7 | — | 75.2 | 1509 | 49.49 | — | — | — |
| 🇨🇳 Qwen3.8-MaxAlibaba · China | 72.6 | 60.6 | 57.5 | 1475 | 69.7 | 91.6 | 81 | 7 |
| 🆕 🇺🇸 GPT-6 AstraOpenAI · USA | 72.5 | — | 73.2 | 1474 | 57.9 | — | 96.0 | — |
| 🆕 🇨🇳 Qwen3.7-PlusAlibaba · China | 68.5 | 57.6 | — | 1455 | 70.3 | — | 90.3 | — |
| 🇨🇳 DeepSeek-V4.1-FlashDeepSeek · China | 66.1 | 55.4 | 62.8 | 1450 | 67.9 | 93.5 | 82 | 9 |
| 🆕 🇺🇸 Gemini 3.5 Flash-LiteGoogle · USA | 65.9 | 54.2 | — | 1455 | — | — | — | 10.3 |
| 🇺🇸 Gemini 4 ArgonGoogle · USA | 65.8 | 43.3 | 11.7 | 1501 | 54.2 | 82.9 | 84 | 15 |
| 🇨🇳 GLM-5.3-FlashZhipu AI · China | 61.5 | — | 63.4 | 1475 | 35.45 | — | 91.2 | — |
| 🆕 🇨🇳 DeepSeek-V4.1-Flash-1017DeepSeek · China | 59.9 | — | — | 1476 | 31.82 | — | 90.9 | — |
| 🆕 🇺🇸 Gemini 3.8 FlashGoogle · USA | 58.6 | — | 73.8 | 1498 | 18.18 | — | 95.3 | — |
| 🆕 🇨🇳 MiniMax-M3MiniMax · China | 58.5 | 59.0 | — | 1441 | 46.1 | — | 92.9 | — |
| 🆕 🇺🇸 Muse Spark 1.3 (preview)Meta · USA | 56.4 | — | — | 1496 | 14.85 | — | — | — |
| 🆕 🇺🇸 Grok 4.7xAI · USA | 54.9 | — | — | 1442 | 38.79 | — | — | — |
| 🆕 🇨🇳 Step 5 PreviewStepFun · China | 54.4 | — | — | 1454 | 31.82 | — | 93.5 | — |
| 🆕 🇺🇸 Claude Haiku 5.5Anthropic · USA | 54.2 | 64.8 | — | 1423 | 36.36 | — | — | — |
| 🇺🇸 Gemini 3.8 FlashGoogle · USA | 53.5 | — | 65.3 | 1489 | 12.42 | — | 95.3 | — |
| 🆕 🇺🇸 MAI-Thinking-1Microsoft · USA | 49.4 | 52.8 | — | — | 46.0 | — | 84.2 | — |
| 🆕 🇪🇺 Mistral Medium 3.5Mistral AI · Europa | 46.7 | — | — | 1426 | 30.3 | — | — | — |
| 🆕 🇪🇺 Mistral Large 4 (preview)Mistral AI · Europa | 43.6 | — | — | 1419 | 27.78 | — | — | — |
| 🆕 🇺🇸 GPT-6 LunaOpenAI · USA | 42.6 | — | — | 1440 | 15.15 | — | — | — |
| 🆕 🇨🇳 Qwen3.8-Omni-FlashAlibaba · China | — | 63.3 | — | — | — | — | 91.0 | — |
| 🆕 🇨🇳 Qwen3.8-FlashAlibaba · China | — | 62.5 | — | — | — | — | 91.7 | — |
| 🆕 🇺🇸 MAI-Code-1.1-FlashMicrosoft · USA | — | — | — | — | — | — | — | — |
| 🆕 🇪🇺 Mistral Small 4Mistral AI · Europa | — | — | — | — | 16.9 | — | — | — |
| 🆕 🇨🇳 Doubao-Seed-2.1-ProByteDance · China | — | — | — | — | — | — | — | — |
| 🆕 🇺🇸 Grok Build 0.1xAI · USA | — | — | — | — | — | — | — | — |
| 🆕 🇺🇸 Muse Spark Open (816B-A42B)Meta · USA | — | — | — | — | 6.57 | — | — | — |
| 🆕 🇨🇳 Kimi K2.7 CodeMoonshot AI · China | — | — | 30.5 | — | — | — | — | — |
| 🆕 🇨🇳 GLM-5.3-FlashXZhipu AI · China | — | — | — | — | — | — | — | — |
| 🆕 🇨🇳 MiniMax-M3.1-FlashMiniMax · China | — | — | — | — | 25.25 | — | — | — |
| 🆕 🇨🇳 Qwen3.7-FlashAlibaba · China | — | — | — | — | — | — | — | — |
| 🆕 🇨🇳 Kimi K3.1 ProMoonshot AI · China | — | — | — | — | 65.7 | — | 94.2 | — |
| 🆕 🇺🇸 Claude Mythos 5.1Anthropic · USA | — | — | 67.4 | — | 60.9 | — | — | — |
| 🆕 🇺🇸 Gemini 3.1 Flash-LiteGoogle · USA | — | — | — | 1433 | — | — | — | — |
| 🆕 🇺🇸 GPT-6.1 TerraOpenAI · USA | — | — | — | — | 50.3 | — | — | — |
| 🆕 🇺🇸 Gemini 3.6 FlashGoogle · USA | — | — | 46.7 | 1483 | — | — | — | — |
| 🆕 🇺🇸 Claude Fable 5.2Anthropic · USA | — | — | — | — | 65.15 | — | — | — |
| 🆕 🇺🇸 Claude Mythos 5.2Anthropic · USA | — | — | — | — | — | — | — | — |
Quality = equal-thirds average of SWE-bench-Pro (SW development) + LMArena (human preference) + Terminal-Bench (agentic capability), the three with reliable sources (Zendoric Quality); if one is missing its weight is shared among those present (at least two; otherwise «—»). LiveCodeBench and GPQA are shown for reference (indicative, may be incomplete) but are NOT in the index; ARC-AGI-2 (arcprize.org) tracks AGI progress: models score VERY low → still far from AGI. %, except LMArena (Elo).
💵 Economics (USD / 1M tokens)
| Model | Input | Cache | Output |
|---|---|---|---|
| 🇨🇳 Kimi K3Moonshot AI · China | $3.0 | $0.3 | $15.0 |
| 🇺🇸 Claude Mythos 5.1 (acceso restringido)Anthropic · USA | $10.0 | $0.25 | $50.0 |
| 🆕 🇺🇸 Gemini 4 CarbonGoogle · USA | $2.0 | $0.1 | $10.0 |
| 🆕 🇺🇸 Gemini 4 ArgonGoogle · USA | $2.0 | $0.1 | $10.0 |
| 🇨🇳 GLM-5.3Zhipu AI · China | $1.4 | $0.26 | $4.4 |
| 🆕 🇺🇸 GPT-6.1 SolOpenAI · USA | $2.0 | $0.1 | $10.0 |
| 🇺🇸 Claude Opus 5.5Anthropic · USA | $4.0 | $0.2 | $20.0 |
| 🇺🇸 Claude Sonnet 5.5Anthropic · USA | until Aug 31, 2026 $2.0 from Sep 1, 2026 $3.0 | until Aug 31, 2026 $0.2 from Sep 1, 2026 $0.3 | until Aug 31, 2026 $10.0 from Sep 1, 2026 $15.0 |
| 🆕 🇺🇸 Muse WonderMeta · USA | $4.0 | $0.2 | $20.0 |
| 🇨🇳 Qwen3.8-MaxAlibaba · China | $2.0 | $0.2 | $6.0 |
| 🆕 🇺🇸 GPT-6 AstraOpenAI · USA | $10.0 | $1.0 | $50.0 |
| 🆕 🇨🇳 Qwen3.7-PlusAlibaba · China | $0.4 | $0.04 | $1.6 |
| 🇨🇳 DeepSeek-V4.1-FlashDeepSeek · China | $0.3 | $0.01 | $1.2 |
| 🆕 🇺🇸 Gemini 3.5 Flash-LiteGoogle · USA | $0.3 | $0.03 | $2.5 |
| 🇺🇸 Gemini 4 ArgonGoogle · USA | $2.0 | $0.1 | $10.0 |
| 🇨🇳 GLM-5.3-FlashZhipu AI · China | $0.15 | $0.03 | $0.5 |
| 🆕 🇨🇳 DeepSeek-V4.1-Flash-1017DeepSeek · China | $0.3 | $0.01 | $1.2 |
| 🆕 🇺🇸 Gemini 3.8 FlashGoogle · USA | $0.75 | $0.07 | $3.75 |
| 🆕 🇨🇳 MiniMax-M3MiniMax · China | $0.3 | $0.06 | $1.2 |
| 🆕 🇺🇸 Muse Spark 1.3 (preview)Meta · USA | $1.25 | $0.15 | $4.25 |
| 🆕 🇺🇸 Grok 4.7xAI · USA | $2.0 | $0.5 | $6.0 |
| 🆕 🇨🇳 Step 5 PreviewStepFun · China | $1.0 | $0.05 | $2.7 |
| 🆕 🇺🇸 Claude Haiku 5.5Anthropic · USA | $0.1 | $0.01 | $0.5 |
| 🇺🇸 Gemini 3.8 FlashGoogle · USA | $0.75 | $0.07 | $3.75 |
| 🆕 🇺🇸 MAI-Thinking-1Microsoft · USA | $2.0 | $0.2 | $8.0 |
| 🆕 🇪🇺 Mistral Medium 3.5Mistral AI · Europa | $1.5 | $0.15 | $7.5 |
| 🆕 🇪🇺 Mistral Large 4 (preview)Mistral AI · Europa | $1.36 | $0.14 | $4.18 |
| 🆕 🇺🇸 GPT-6 LunaOpenAI · USA | $0.1 | $0.01 | $0.5 |
| 🆕 🇨🇳 Qwen3.8-Omni-FlashAlibaba · China | $0.15 | $0.02 | $0.47 |
| 🆕 🇨🇳 Qwen3.8-FlashAlibaba · China | $0.15 | $0.01 | $0.47 |
| 🆕 🇺🇸 MAI-Code-1.1-FlashMicrosoft · USA | $0.2 | $0.02 | $1.2 |
| 🆕 🇪🇺 Mistral Small 4Mistral AI · Europa | $0.15 | $0.01 | $0.6 |
| 🆕 🇨🇳 Doubao-Seed-2.1-ProByteDance · China | $0.84 | $0.17 | $4.21 |
| 🆕 🇺🇸 Grok Build 0.1xAI · USA | $1.0 | $0.2 | $2.0 |
| 🆕 🇺🇸 Muse Spark Open (816B-A42B)Meta · USA | $1.25 | $0.15 | $4.25 |
| 🆕 🇨🇳 Kimi K2.7 CodeMoonshot AI · China | $0.95 | $0.19 | $4.0 |
| 🆕 🇨🇳 GLM-5.3-FlashXZhipu AI · China | $0.37 | $0.07 | $1.25 |
| 🆕 🇨🇳 MiniMax-M3.1-FlashMiniMax · China | $0.15 | $0.01 | $0.6 |
| 🆕 🇨🇳 Qwen3.7-FlashAlibaba · China | $0.03 | $0.0 | $0.13 |
| 🆕 🇨🇳 Kimi K3.1 ProMoonshot AI · China | $2.0 | $0.2 | $10.0 |
| 🆕 🇺🇸 Claude Mythos 5.1Anthropic · USA | $10.0 | $0.25 | $50.0 |
| 🆕 🇺🇸 Gemini 3.1 Flash-LiteGoogle · USA | $0.25 | $0.03 | $1.5 |
| 🆕 🇺🇸 GPT-6.1 TerraOpenAI · USA | $1.0 | $0.05 | $5.0 |
| 🆕 🇺🇸 Gemini 3.6 FlashGoogle · USA | $1.5 | $0.07 | $7.5 |
| 🆕 🇺🇸 Claude Fable 5.2Anthropic · USA | $8.0 | $0.2 | $40.0 |
| 🆕 🇺🇸 Claude Mythos 5.2Anthropic · USA | $8.0 | $0.2 | $40.0 |
Claude Sonnet 5: scheduled price increase (same model) — reduced pricing until Aug 31, 2026 and standard pricing from Sep 1, 2026.
🔓 Open source & type
| Model | Open source | License | Type |
|---|---|---|---|
| 🇨🇳 Kimi K3Moonshot AI · China | Yes | Kimi K3 License | Open-weight |
| 🇺🇸 Claude Mythos 5.1 (acceso restringido)Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Gemini 4 CarbonGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Gemini 4 ArgonGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🇨🇳 GLM-5.3Zhipu AI · China | Yes | GLM-5.3 (propia) | Open-weight |
| 🆕 🇺🇸 GPT-6.1 SolOpenAI · USA | No | Proprietary | Proprietary (API only) |
| 🇺🇸 Claude Opus 5.5Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🇺🇸 Claude Sonnet 5.5Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Muse WonderMeta · USA | No | Proprietary | Proprietary (API only) |
| 🇨🇳 Qwen3.8-MaxAlibaba · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 GPT-6 AstraOpenAI · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 Qwen3.7-PlusAlibaba · China | No | Proprietary | Proprietary (API only) |
| 🇨🇳 DeepSeek-V4.1-FlashDeepSeek · China | Yes | MIT | Open-weight |
| 🆕 🇺🇸 Gemini 3.5 Flash-LiteGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🇺🇸 Gemini 4 ArgonGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🇨🇳 GLM-5.3-FlashZhipu AI · China | Yes | MIT | Open-weight |
| 🆕 🇨🇳 DeepSeek-V4.1-Flash-1017DeepSeek · China | Yes | MIT | Open-weight |
| 🆕 🇺🇸 Gemini 3.8 FlashGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 MiniMax-M3MiniMax · China | Yes | MiniMax Community License | Open-weight |
| 🆕 🇺🇸 Muse Spark 1.3 (preview)Meta · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Grok 4.7xAI · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 Step 5 PreviewStepFun · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Claude Haiku 5.5Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🇺🇸 Gemini 3.8 FlashGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 MAI-Thinking-1Microsoft · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇪🇺 Mistral Medium 3.5Mistral AI · Europa | Yes | Modified MIT | Open-weight |
| 🆕 🇪🇺 Mistral Large 4 (preview)Mistral AI · Europa | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 GPT-6 LunaOpenAI · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 Qwen3.8-Omni-FlashAlibaba · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 Qwen3.8-FlashAlibaba · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 MAI-Code-1.1-FlashMicrosoft · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇪🇺 Mistral Small 4Mistral AI · Europa | Yes | Apache-2.0 | Open-weight |
| 🆕 🇨🇳 Doubao-Seed-2.1-ProByteDance · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Grok Build 0.1xAI · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Muse Spark Open (816B-A42B)Meta · USA | Yes | Muse Model License | Open-weight |
| 🆕 🇨🇳 Kimi K2.7 CodeMoonshot AI · China | Yes | Modified MIT | Open-weight |
| 🆕 🇨🇳 GLM-5.3-FlashXZhipu AI · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 MiniMax-M3.1-FlashMiniMax · China | Yes | MiniMax Community License | Open-weight |
| 🆕 🇨🇳 Qwen3.7-FlashAlibaba · China | No | Proprietary | Proprietary (API only) |
| 🆕 🇨🇳 Kimi K3.1 ProMoonshot AI · China | Yes | — | Open-weight |
| 🆕 🇺🇸 Claude Mythos 5.1Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Gemini 3.1 Flash-LiteGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 GPT-6.1 TerraOpenAI · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Gemini 3.6 FlashGoogle · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Claude Fable 5.2Anthropic · USA | No | Proprietary | Proprietary (API only) |
| 🆕 🇺🇸 Claude Mythos 5.2Anthropic · USA | No | Proprietary | Proprietary (API only) |
🖥️ Open source you can self-host
Small/medium models you can run on your own machine (laptop/PC/Mac). Quality = Artificial Analysis Intelligence Index (0-100; output quality), the measure with best coverage of small open models (LMArena does not list sub-32B). Memory estimated at 4-bit (Q4) and 8-bit (Q8) quantization; on Apple Silicon it is UNIFIED memory (RAM=VRAM).
| Model | Quality (AA Index) | GPQA | Params | RAM Q4 | RAM Q8 | GPU | CPU / Mac | License |
|---|---|---|---|---|---|---|---|---|
| Gemma 4 31BGoogle | 39 | 84.3 | 31B | 19 GB | 36 GB | ≥24 GB | Limitado (mejor GPU/Mac ≥32 GB) | Gemma |
| Gemma 4 26B A4BGoogle | 31 | 82.3 | 26B | 16 GB | 30 GB | ≥16 GB | Limitado (mejor GPU/Mac ≥32 GB) | Gemma |
| Nemotron Cascade 2 30B A3BNVIDIA | 28 | 76.1 | 30B | 18 GB | 34 GB | ≥24 GB | Limitado (mejor GPU/Mac ≥32 GB) | NVIDIA Open Model |
| gpt-oss-20bOpenAI | 24 | 71.5 | 20B | 13 GB | 25 GB | ≥16 GB | Limitado (mejor GPU/Mac ≥32 GB) | Apache-2.0 |
| Gemma 4 12BGoogle | 22 | 78.8 | 12B | 8 GB | 15 GB | ≥8 GB | Sí (CPU lento · Mac 16 GB) | Gemma |
| 🆕 🇨🇳 Qwen3.8-27BAlibaba | 20 | 89.2 | 27B | 16 GB | 31 GB | ≥16 GB | Limitado (mejor GPU/Mac ≥32 GB) | Apache-2.0 |
| Gemma 4 E4BGoogle | 19 | 58.6 | 4B | 4 GB | 6 GB | ≥8 GB | Sí (CPU/Mac, fluido) | Gemma |
| 🆕 🇨🇳 Qwen3.6-35B-A3BAlibaba | 18 | 86.0 | 35B | 21 GB | 40 GB | ≥24 GB | Limitado (mejor GPU/Mac ≥32 GB) | Apache-2.0 |
| Gemma 4 E2BGoogle | 15 | 43.4 | 2B | 3 GB | 4 GB | ≥8 GB | Sí (CPU/Mac, fluido) | Gemma |
| 🆕 Muse Glimmer 30BMeta | 6.3 | — | 29.6B | 18 GB | 34 GB | ≥24 GB | Limitado (mejor GPU/Mac ≥32 GB) | Apache-2.0 |
| 🆕 🇪🇺 Ministral 3 14BMistral AI | 6 | 71.2 | 14B | 9 GB | 17 GB | ≥12 GB | Sí (CPU lento · Mac 16 GB) | Apache-2.0 |
| 🆕 🇪🇺 Ministral 3 8B (25.12)Mistral AI | 5 | — | 8B | 6 GB | 10 GB | ≥8 GB | Sí (CPU/Mac, fluido) | Apache-2.0 |
| 🆕 🇪🇺 Ministral 3 3B (25.12)Mistral AI | 5 | — | 3B | 3 GB | 5 GB | ≥8 GB | Sí (CPU/Mac, fluido) | Apache-2.0 |
🗄️ Large open source (server / multi-GPU)
Powerful open models that need a server or multiple GPUs. Quality = LMArena Elo (human preference over output, source lmarena.ai), which does cover large models. For MoE, memory counts total parameters (all experts are loaded). Memory estimated at 4-bit (Q4) and 8-bit (Q8) quantization; on Apple Silicon it is UNIFIED memory (RAM=VRAM).
| Model | Quality (LMArena) | GPQA | Params | RAM Q4 | RAM Q8 | GPU | CPU / Mac | License |
|---|---|---|---|---|---|---|---|---|
| 🆕 Kimi K3Moonshot AI | 1487 | 93.5 | 2800B | 1542 GB | 3082 GB | 20× 80 GB (servidor) | No (servidor GPU) | Kimi K3 License |
| 🆕 GLM-5.3Zhipu AI | 1483 | — | 753B | 416 GB | 830 GB | 6× 80 GB (servidor) | No (servidor GPU) | glm-5.3 (custom) |
| 🆕 Qwen3.8-2.4T-A95BAlibaba | 1482 | 92.6 | 2400B | 1322 GB | 2642 GB | 17× 80 GB (servidor) | No (servidor GPU) | Qwen3.8-Max License |
| 🆕 DeepSeek-V4.1-Flash-1017DeepSeek | 1476 | 90.9 | 552B | 305 GB | 609 GB | 4× 80 GB (servidor) | No (servidor GPU) | MIT |
| Mistral Large 3Mistral AI | 1416 | 68.0 | 675B | 373 GB | 744 GB | 5× 80 GB (servidor) | No (servidor GPU) | Apache-2.0 |
| gpt-oss-120bOpenAI | 1366 | 80.1 | 117B | 66 GB | 130 GB | ≥80 GB | No (servidor GPU) | Apache-2.0 |
| 🆕 Hy4 (preview)Tencent | — | 92.3 | 780B | 431 GB | 860 GB | 6× 80 GB (servidor) | No (servidor GPU) | Apache-2.0 |
🎨 Image generation
The leading IMAGE generation models, split between remote services (pay per image via API) and open models you can run on your own hardware.
☁️ Remote (API / cloud)
| Model | Quality (arena) | Price (per image) | Notes |
|---|---|---|---|
| GPT Image 2OpenAI · USA | AA t2i v2.1: 1188, ≥nº5 (2.6 Sunburst 1259 nº1, 2.6 Flare 1228, 2.5 Sunburst 1227). AA edit 1133 nº6. arena.ai 1381 nº3 | $0,211/img high 1024² (OpenAI $30/M tok img out); fal.ai $0,006/img low 1024² a $0,401/img high 4K | Lanz. 21-abr-26. Superado por 2.5 (8-sep-26) y 2.6 Flare/Sunburst ($30/M tok img out; ~$0,21/img max 1024² según terceros). Azure retira 21-oct-27 |
| MAI-Image-2.5Microsoft · USA | AA t2i v2.1: 1118 nº12/106 (GPT Image 1.5 1123 nº9; MAI-2.6 1167 nº6). AA edit 1125 nº9. arena.ai 1253 nº13 (26-oct-26) | Azure Foundry $47/M tok img out ($5 txt in, $8 img in); AA ~$0,048/img; fal.ai ~$0,05/img | Retirado de Foundry 1-oct-26. Sucesores: MAI-2.6 ($38/M tok, ~$0,039/img, preview), 2.6-Flash ($19/M tok img out); 2.7 Preview sin API |
| Nano Banana Pro (Gemini 3 Pro Image)Google · USA | AA t2i v2.1: 1118 nº13/106; superada por Nano Banana 2.1 (1176 nº4). AA edit v2.1 1109 nº14/69. arena.ai 2K 1246 nº16 | Gemini API: $0,134/img 1-2K · $0,24/img 4K (Batch: $0,067 / $0,12/img) | = Gemini 3 Pro Image. GA 28-may-26; preview apagado 25-jun-26. Sin retiro. Nano Banana 2.1: $0,0336/img 1K, $0,113/img 4K |
| Nano Banana 2 Lite (Gemini 3.1 Flash Lite)Google · USA | AA t2i v2.1: 1107 nº15/106; AA edit v2.1 1050 nº27/69; arena.ai 1249 nº15 (26-oct-26) | Gemini API $0,0336/img (1K); Batch $0,0168/img | = Gemini 3.1 Flash Lite Image. GA 30-jun-26. Sin retiro; sustituye a 2.5 Flash Image (retiro 15-mar-2027) |
| GPT Image 1.5OpenAI · USA | AA t2i v2.1: 1123 nº9/106, sobre MAI-Image-2.5 (1118 nº12). AA edit v2.1 1116 nº12. arena.ai High Fidelity 1236 nº17 | $0,009/img low · $0,034 medio · $0,133 high (1024²; $32/M tok img out, OpenAI) | RETIRO OpenAI 1-dic-2026 (aviso 2-jun-26) con gpt-image-1-mini y chatgpt-image-latest; Azure 16-dic-26. Reemplazo oficial: gpt-image-2 |
| Grok ImaginexAI · USA | AA t2i v2.1: 2.0 nº5 (1168); quality nº20 (1049); base nº34 (1022). AA edit 2.0 nº13 (1112). arena.ai 2.0 nº6 (1317) | x.ai: image $0,02/img · image-2.0 $0,04/img low 1K (quality $0,05, retirado); vídeo 1.5 $0,08/s, Lite $0,02/s | image-quality y alias -pro RETIRADOS 2-nov-26 (docs.x.ai): redirigen a image-2.0 low ($0,01/img menos). Batch solo admite image 1.0 |
| Cosmos 3 SuperNVIDIA · USA | AA t2i v2.1: agentic 993 (nº50), base 986 (nº58), 4-step 971 (nº65) de 106. arena.ai agentic 1164 #27, base 1151 #34 | Pesos abiertos OpenMDW-1.1 (self-host gratis); fal $0,04/img (+$0,02 prompt expansion), i2v $0,05/s | NVIDIA, pesos 31-may-26. MoT 64B omnimodal. No es nº1 open en AA v2.1: Qwen-Image-2.1 1046, FLUX 3 Klein 1042, Fibo 2.0 1040 |
💻 Local (open-weight, on your hardware)
| Model | License | Hardware | Notes |
|---|---|---|---|
| FLUX.2Black Forest Labs | open-weight + API | GPU dedicada (variantes dev/FP8) | Lanz. 25-nov-25; klein 15-ene-26. Superado por FLUX 3 Image (1-oct-26, cerrado, $0,048/img 1K); FLUX 3 Klein 1042, nº2 open AA |
| Qwen-Image 2.0Alibaba | Apache-2.0 | 7B — GPU 16GB+ | Lanz. 10-feb-26, 2K nativo. 3.0 (21-jul-26) cerrado sin pesos. Último abierto: 2.1 (20-sep-26, lic. research) |
| Z-Image TurboAlibaba (Tongyi) | open-weight | ~1 s/imagen en H100; corre en Mac (MLX) | Alibaba Tongyi-MAI, 6B S3-DiT, 8 pasos. Turbo 26-nov-2025, Base 27-ene-2026; Edit y Omni-Base aún sin publicar |
| Stable Diffusion 3.5Stability AI | Stability Community | GPU 8-16GB | Sigue 'current' en Stability, no retirado; Azure lo retiró 31-jul-2026. Flagship actual: Stable Image Ultra $0,08/img |
| HiDream-O1-Image-1.5HiDream | open-weight | GPU dedicada | v1.5 cerrada (jun-2026), 2K pixel-native sin VAE. Base/Dev MIT 8-may-2026. Debutó nº3 en AA, hoy nº39 |
Data as of 2026-11-12 · sources: Artificial Analysis Image/Video Arena · arena.ai · llm-stats.com · Pixazo · documentación de fabricantes · arena Elo is blind human preference over the output; indicative prices — verify before deciding.
🎬 Video generation
The leading VIDEO generation models, split between remote services (pay per generated second via API) and open models for your own hardware.
☁️ Remote (API / cloud)
| Model | Quality (arena) | Price (per second) | Notes |
|---|---|---|---|
| Kling 3.0 / TurboKuaishou · China | AA t2v v2.1: Omni 1080p 1021 #18, 3.0 Pro 1000 #19/31; i2v v2.1: Omni 1002, Pro 1000 (#17-19/29). arena.ai i2v 1354 #19 | Turbo con audio $0,14/s 1080p, $0,112/s 720p; 3.0 sin audio $0,112/$0,084/s, 4K $0,42/s. fal v3 Pro audio $0,168/s | Kling 4.0 (15-oct-26) en API ~28-oct: $0,157/s 720p, $0,235/s 1080p (Kling/fal/WaveSpeed), clips 3-30s. Sin Elo en AA |
| Veo 3.1Google · USA | AA t2v v2.1: 3.1 961 #22, Fast 957 #23, Lite 951 #24/31. AA i2v: 3.1 955 #21, Fast 936 #22, Lite 929 #24. arena.ai t2v 1364 #12 | Gemini API: Lite $0,05/s 720p ($0,08 1080p) · Fast $0,10/s ($0,12 1080p, $0,30 4K) · Std $0,40/s, $0,60/s 4K | veo-3.1-*-preview apagados en Gemini API 22-oct-26 → gemini-omni-1.1-flash. GA sigue en Flow/Enterprise. fal depreca 17-nov-26 |
| Gemini Omni FlashGoogle · USA | AA t2v v2.1: 1.1 1129 #10/31. AA i2v v2.1 1043 #14/29. arena.ai (8-nov-26): 1.2 Flash t2v 1540 #1, i2v 1503 #1; 1.1 t2v 1513 #2 | 1.1: $0,10/s 720p · 360p $0,03/s · 1080p $0,15/s · 4K $0,30/s. 1.2 Flash: $0,07/s 720p (-30%) | Gemini Omni 1.2 Flash lanzado 4/5-nov-26 (clips 15s); arena.ai lo marca preliminar. Desde 3-dic-26 omni-1.1-flash se enruta a 1.2 |
| Seedance 2.0 (Dreamina)ByteDance · China | AA t2v v2.1: 1136 #9/31 (2.5 = 1155 #5). AA i2v v2.1: 1135 #3/29 (2.5 = 1159 #1). arena.ai t2v 1483 #6 (2.5: 1479 #7, 8-nov-26) | Dreamina promo expirada 9-oct-26. Super anual: 2.0 $0,086/s, Fast $0,041/s, 2.5 $0,047/s. BytePlus 2.5: $10,70/M tok | ByteDance 12-feb-26, 15s+audio. Superado por Seedance 2.5 (31-jul-26): 30s, 50 refs. API 2.5 BytePlus ≤1080p; 4K solo enterprise |
| Sora 2OpenAI · USA | Fuera de AA t2v v2.1 y AA i2v v2.1. arena.ai: Pro 1368 #11, Sora 2 1343 #17 (21-sep-2026) | RETIRADO: sin precio vigente. Histórico $0,10/s 720p; Pro $0,30-0,70/s | OpenAI: app cerrada 26-abr-26; Videos API y alias sora-2/-pro apagados 24-sep-26; Azure preview retirado 15-oct-26. Sin reemplazo |
💻 Local (open-weight, on your hardware)
| Model | License | Hardware | Notes |
|---|---|---|---|
| Wan 2.7Alibaba | open-weight | 16-24GB VRAM (14B) | Alibaba, 7-abr-26. T2V/I2V/ref-to-video + edición, audio nativo, 2-15s. Superado por Wan 3.0 (GA 24-ago-26, nº1 AA t2v, nº5 i2v) |
| LTX-2.3Lightricks | open-weight | desde 12GB VRAM (FP8: flujos de 32GB) | Lightricks, 22B DiT, audio+vídeo, 5-mar-2026, hasta 4K/20s. Sucesor LTX-2.5 (11-ago-2026), pago si ≥$10M/año |
| HunyuanVideoTencent | open-weight | 16-24GB VRAM | Tencent 13B, open 3-dic-2024. Sucesor HunyuanVideo 1.5 (8,3B, ~14GB VRAM, 20-nov-2025) |
| Mochi 1Genmo | open-weight | fine-tune en 1×H100/A100 80GB | Genmo, 10B AsymmDiT, 22-oct-2024, Apache 2.0, 480p. Legacy en 2026 (superado por Wan 2.2, Hunyuan 1.5, MiniMax H3) |
| SkyReels V1Skywork | open-weight | GPU dedicada | Skywork, 18-feb-2025, fine-tune de HunyuanVideo. Superado: V2 abr-2025, V3 29-ene-2026, V4 25-feb-2026 |
Data as of 2026-11-12 · sources: Artificial Analysis Image/Video Arena · arena.ai · llm-stats.com · Pixazo · documentación de fabricantes · arena Elo is blind human preference over the output; indicative prices — verify before deciding.