{"slug": "vals-ai-compare-ai-models-across-real-world-benchmarks", "title": "Vals AI – Compare AI models across real-world benchmarks", "summary": "Vals AI's benchmark comparison page, dated 2026-11-18, ranks Claude Fable 5.2 first on the GDP-weighted Vals Index at 71.31% (±0.96), ahead of Gemini 4 Neon at 70.87% (±0.94), Claude Sonnet 5.5 at 70.85% (±0.93), Gemini 4 Carbon at 70.45% (±0.97) and Claude Opus 5.5 at 70.14% (±1.00). The index aggregates tasks across finance, coding, legal and tax, and the same page shows Gemini 4 Neon leading the Legal Research Bench at 56.25% (±3.45) and Tax Agent Bench at 54.89% (±3.59), while Claude Fable 5.2 posts the highest Vibe Code Bench v1.1 score at 93.77% (±1.15). Vals AI lists per-task Vals Index costs of $17.94 for Claude Fable 5.2, $7.405 for Gemini 4 Neon, $16.18 for Claude Sonnet 5.5, $10.51 for Gemini 4 Carbon and $24.77 for Claude Opus 5.5.", "body_md": "# Compare Models\n\nSelect models (max 7)\n\nClaude Fable 5.2Gemini 4 NeonClaude Sonnet 5.5Gemini 4 CarbonClaude Opus 5.5\n\n| Vals IndexGDP-weighted benchmark | 71.31%±0.96 | 70.87%±0.94 | 70.85%±0.93 | 70.45%±0.97 | 70.14%±1.00 | \n| Legal Research BenchAgentic US legal research | 46.15%±3.46 | 56.25%±3.45 | 48.08%±3.47 | 55.77%±3.45 | 50.48%±3.48 | \n| Finance Agent v2.1Core financial analyst tasks | 69.47%±1.98 | 68.56%±1.94 | 67.28%±2.02 | 67.16%±2.02 | 67.85%±2.05 | \n| Tax Agent BenchAgentic US corporate tax research | 32.09%±3.37 | 54.89%±3.59 | 41.96%±3.56 | 51.78%±3.60 | 44.54%±3.58 | \n| MedCodeMedical billing code support | N/A | N/A | 52.92%±2.12 | N/A | 49.80%±2.27 | \n| Terminal-Bench-ScienceExpert-authored scientific research workflows | 55.71%±5.98 | 62.86%±5.82 | 47.14%±6.01 | N/A | 47.14%±6.01 | \n| Code Migration v1.1Rewriting programs in new languages | 83.27%±3.12 | N/A | 86.88%±1.52 | N/A | 81.01%±3.21 | \n| Terminal-Bench 4.0Frontier-difficulty terminal tasks | 65.15%±2.31 | 61.62%±2.02 | 64.14%±1.01 | 63.13%±2.20 | 65.15%±0.00 | \n| Vibe Code Bench v1.1Building web apps from scratch | 93.77%±1.15 | 93.70%±1.17 | 92.39%±1.26 | 89.06%±2.29 | 90.29%±1.53 | \n\nBenchmarks\n\nVals Index *\n\nClaude Fable 5.2\n\n0.00%± 0.96\n\n(42/42)\nGemini 4 Neon\n\n0.00%± 0.94\n\n(42/42)\nClaude Sonnet 5.5\n\n0.00%± 0.93\n\n(42/42)\nGemini 4 Carbon\n\n0.00%± 0.97\n\n(42/42)\nClaude Opus 5.5\n\n0.00%± 1.00\n\n(42/42)\nLegal Research Bench *\n\nClaude Fable 5.2\n\n0.00%± 3.46\n\n(99/99)\nGemini 4 Neon\n\n0.00%± 3.45\n\n(99/99)\nClaude Sonnet 5.5\n\n0.00%± 3.47\n\n(99/99)\nGemini 4 Carbon\n\n0.00%± 3.45\n\n(99/99)\nClaude Opus 5.5\n\n0.00%± 3.48\n\n(99/99)\nFinance Agent v2.1 *\n\nClaude Fable 5.2\n\n0.00%± 1.98\n\n(45/45)\nGemini 4 Neon\n\n0.00%± 1.94\n\n(45/45)\nClaude Sonnet 5.5\n\n0.00%± 2.02\n\n(45/45)\nGemini 4 Carbon\n\n0.00%± 2.02\n\n(45/45)\nClaude Opus 5.5\n\n0.00%± 2.05\n\n(45/45)\nTax Agent Bench *\n\nClaude Fable 5.2\n\n0.00%± 3.37\n\n(90/90)\nGemini 4 Neon\n\n0.00%± 3.59\n\n(90/90)\nClaude Sonnet 5.5\n\n0.00%± 3.56\n\n(90/90)\nGemini 4 Carbon\n\n0.00%± 3.60\n\n(90/90)\nClaude Opus 5.5\n\n0.00%± 3.58\n\n(90/90)\nMedCode *\n\nClaude Fable 5.2\n\nN/A\nGemini 4 Neon\n\nN/A\nClaude Sonnet 5.5\n\n0.00%± 2.12\n\n(108/108)\nGemini 4 Carbon\n\nN/A\nClaude Opus 5.5\n\n0.00%± 2.27\n\n(108/108)\nTerminal-Bench-Science\n\nClaude Fable 5.2\n\n0.00%± 5.98\n\n(48/48)\nGemini 4 Neon\n\n0.00%± 5.82\n\n(48/48)\nClaude Sonnet 5.5\n\n0.00%± 6.01\n\n(48/48)\nGemini 4 Carbon\n\nN/A\nClaude Opus 5.5\n\n0.00%± 6.01\n\n(48/48)\nCode Migration v1.1 *\n\nClaude Fable 5.2\n\n0.00%± 3.12\n\n(30/30)\nGemini 4 Neon\n\nN/A\nClaude Sonnet 5.5\n\n0.00%± 1.52\n\n(30/30)\nGemini 4 Carbon\n\nN/A\nClaude Opus 5.5\n\n0.00%± 3.21\n\n(30/30)\nTerminal-Bench 4.0\n\nClaude Fable 5.2\n\n0.00%± 2.31\n\n(69/69)\nGemini 4 Neon\n\n0.00%± 2.02\n\n(69/69)\nClaude Sonnet 5.5\n\n0.00%± 1.01\n\n(69/69)\nGemini 4 Carbon\n\n0.00%± 2.20\n\n(69/69)\nClaude Opus 5.5\n\n0.00%± 0.00\n\n(69/69)\nVibe Code Bench v1.1 *\n\nClaude Fable 5.2\n\n0.00%± 1.15\n\n(133/133)\nGemini 4 Neon\n\n0.00%± 1.17\n\n(133/133)\nClaude Sonnet 5.5\n\n0.00%± 1.26\n\n(133/133)\nGemini 4 Carbon\n\n0.00%± 2.29\n\n(133/133)\nClaude Opus 5.5\n\n0.00%± 1.53\n\n(133/133)\n## Overall performance\n\nPerformance on the Vals Index, a GDP-weighted aggregation of tasks across finance, coding, legal, and tax\n\n•2026-11-18\n\n## Comparison by Industry\n\nModel performance on different sections of the economy.\n\nLegal\n\n41.71%\n\nvs.\n37.92%\n\nvs.\n42.49%\n\nvs.\n36.63%\n\nvs.\n43.93%\n\nFinance\n\n59.65%\n\nvs.\n66.64%\n\nvs.\n71.28%\n\nvs.\n65.14%\n\nvs.\n72.05%\n\nHealthcare\n\nN/A\n\nvs.\nN/A\n\nvs.\n72.01%\n\nvs.\nN/A\n\nvs.\n70.61%\n\nMath\n\n92.86%\n\nvs.\nN/A\n\nvs.\n100.00%\n\nvs.\n71.43%\n\nvs.\n100.00%\n\nScience\n\n63.74%\n\nvs.\n62.63%\n\nvs.\n59.12%\n\nvs.\nN/A\n\nvs.\n47.63%\n\nEducation\n\nN/A\n\nvs.\nN/A\n\nvs.\n88.03%\n\nvs.\nN/A\n\nvs.\n91.30%\n\nCoding\n\n64.70%\n\nvs.\n77.66%\n\nvs.\n65.22%\n\nvs.\n67.31%\n\nvs.\n64.45%\n\nCyber\n\nN/A\n\nvs.\nN/A\n\nvs.\n44.87%\n\nvs.\nN/A\n\nvs.\n55.35%\n\nSocial Mobility\n\nN/A\n\nvs.\nN/A\n\nvs.\n72.36%\n\nvs.\nN/A\n\nvs.\n73.52%\n\n## Cost Analysis\n\nPer task and per token model pricing.\n\nVals Index · USD\n\n$17.94\n\n$7.405\n\n$16.18\n\n$10.51\n\n$24.77\n\nCost / TestVals Index\n\n$17.94\n\nvs.\n$7.405\n\nvs.\n$16.18\n\nvs.\n$10.51\n\nvs.\n$24.77\n\nInput Cost/ 1M Tokens\n\n$8.00\n\nvs.\n$3.00\n\nvs.\n$2.00\n\nvs.\n$4.00\n\nvs.\n$4.00\n\nInput Cache Write/ 1M Tokens\n\n$10.00\n\nvs.\nN/A\n\nvs.\n$2.50\n\nvs.\nN/A\n\nvs.\n$5.00\n\nInput Cache Read/ 1M Tokens\n\n$0.20\n\nvs.\n$0.15\n\nvs.\n$0.10\n\nvs.\n$0.20\n\nvs.\n$0.20\n\nOutput Cost/ 1M Tokens\n\n$40.00\n\nvs.\n$15.00\n\nvs.\n$10.00\n\nvs.\n$20.00\n\nvs.\n$20.00\n\n## Model Metadata\n\nBasic information about each model.\n\n| Model provider | Anthropic | Google | Anthropic | Google | Anthropic | \n| Latency | 2621.82s | 1884.68s | 5158.77s | 2864.86s | 5071.09s | \n| Cost (In/Out) | $8 / $40 | $3 / $15 | $2 / $10 | $4 / $20 | $4 / $20 | \n| Context Window | 1M | 1M | 1M | 1M | 1M | \n| Max Output Token | 128k | 1M | 128k | 1M | 128k | \n| Input Modality |  |  |  |  |  |", "url": "https://wpnews.pro/news/vals-ai-compare-ai-models-across-real-world-benchmarks", "canonical_source": "https://www.vals.ai/comparison", "published_at": "2026-11-19 05:28:34+00:00", "updated_at": "2026-11-19 05:45:05.306544+00:00", "lang": "en", "topics": ["artificial-intelligence", "large-language-models", "ai-research", "ai-agents"], "entities": ["Vals AI", "Claude Fable 5.2", "Gemini 4 Neon", "Claude Sonnet 5.5", "Gemini 4 Carbon", "Claude Opus 5.5", "Vals Index", "Terminal-Bench 4.0"], "also_reported_by": [], "takeaway": "On Vals AI's GDP-weighted Vals Index, Claude Fable 5.2 leads at 71.31% (±0.96) versus 70.87% for Gemini 4 Neon, but Gemini 4 Neon wins Legal Research Bench (56.25% vs 46.15%) and Tax Agent Bench (54.89% vs 32.09%), so model choice should follow the task domain rather than the aggregate score.", "recommendation": {"advice": "Evaluate models per task domain rather than by aggregate index: Gemini 4 Neon beat Claude Fable 5.2 by 10.1 points on Legal Research Bench and 22.8 points on Tax Agent Bench despite trailing on the overall Vals Index.", "playbook": "evals-testing", "playbook_url": "https://wpnews.pro/playbooks/evals-testing"}, "alternates": {"html": "https://wpnews.pro/news/vals-ai-compare-ai-models-across-real-world-benchmarks", "markdown": "https://wpnews.pro/news/vals-ai-compare-ai-models-across-real-world-benchmarks.md", "text": "https://wpnews.pro/news/vals-ai-compare-ai-models-across-real-world-benchmarks.txt", "jsonld": "https://wpnews.pro/news/vals-ai-compare-ai-models-across-real-world-benchmarks.jsonld"}}