Claude 4.5 Haiku (Reasoning) vs Trinity Large Thinking

Mana yang lebih baik untuk uang: Claude 4.5 Haiku (Reasoning) oleh Anthropic atau Trinity Large Thinking oleh Arcee AI? Kecerdasan, total biaya benchmark, dan waktu end-to-end dari Artificial Analysis & sumber lain, cuplikan 2026-08-19.

Kompromi
AnthropicClaude 4.5 Haiku (Reasoning)Di luar batas Pareto
Skor kecerdasan (Intelligence Index)
29.90
Biaya benchmark (USD)
$471
Arcee AITrinity Large ThinkingDi luar batas Pareto
Skor kecerdasan (Intelligence Index)
18.70
Biaya benchmark (USD)
$255
Kesimpulan

Tergantung anggaran Anda

Claude 4.5 Haiku (Reasoning) mendapat 11.2 poin lebih tinggi (29.9 vs 18.7), tetapi biaya menjalankan benchmark-nya 1.8× lebih besar ($471 vs $255; keduanya kelas menengah). Pilih Claude 4.5 Haiku (Reasoning) jika kecerdasan tambahan sepadan dengan biaya lebihnya; jika tidak, Trinity Large Thinking.

MetrikClaude 4.5 Haiku (Reasoning)Trinity Large ThinkingSelisih
Kecerdasan (lebih tinggi lebih baik)29.918.7−37.5% Trinity Large Thinking 37.5% kurang cerdas
Total biaya AA (lebih rendah lebih baik)$471.22$254.72−45.9% Trinity Large Thinking 45.9% lebih murah
Waktu ujung ke ujung (lebih rendah lebih baik)22.28s14.08s−36.8% Trinity Large Thinking 36.8% lebih cepat
Harga per 1 juta token (masuk/keluar)$1.00 / $5.00$0.23 / $0.88
Jenis inputTidak dilaporkanTidak dilaporkan

Trinity Large Thinking dibanding Claude 4.5 Haiku (Reasoning): kecerdasan 37.5% lebih rendah, biaya menjalankan benchmark AA 45.9% lebih rendah, dan waktu total 36.8% lebih cepat.

Buka pertarungan ini di pemilih interaktifClaude 4.5 Haiku (Reasoning): model apa yang mengunggulinyaTrinity Large Thinking: model apa yang mengunggulinya
Pareto frontier hari ini

Dibandingkan dengan batas Pareto

0204060$10$20$50$100$200$500$1k$2k$5kBiaya untuk menjalankan Artificial Analysis Intelligence Index (USD, skala log)Indeks kecerdasanCeleris-1 (Celeris): kecerdasan 12.40, $157Claude 4.5 Sonnet (Reasoning) (Anthropic): kecerdasan 37.40, $1037Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) (Anthropic): kecerdasan 62.10, $5455Claude Opus 4.7 (Adaptive Reasoning, Max Effort) (Anthropic): kecerdasan 55.00, $3743Claude Opus 4.8 (Adaptive Reasoning, Max Effort) (Anthropic): kecerdasan 57.30, $3752Claude Opus 5 (Adaptive Reasoning, Low Effort) (Anthropic): kecerdasan 52.50, $555Claude Opus 5 (Adaptive Reasoning, Medium Effort) (Anthropic): kecerdasan 58.60, $1116Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) (Anthropic): kecerdasan 48.40, $3355Claude Sonnet 5 (Adaptive Reasoning, Max Effort) (Anthropic): kecerdasan 55.30, $4011Claude Sonnet 5 (Non-reasoning, High Effort) (Anthropic): kecerdasan 42.60, $416DeepSeek R1 (Jan '25) (DeepSeek): kecerdasan 18.60, $417DeepSeek V3 (Dec '24) (DeepSeek): kecerdasan 14.20, $20.0DeepSeek V3 0324 (DeepSeek): kecerdasan 15.20, $35.2DeepSeek V4 Flash (Reasoning, High Effort) (DeepSeek): kecerdasan 39.00, $57.0DeepSeek V4 Flash (Reasoning, Max Effort) (DeepSeek): kecerdasan 42.10, $104DeepSeek V4 Flash 0731 (Reasoning, Max Effort) (DeepSeek): kecerdasan 51.80, $323DeepSeek V4 Pro (Reasoning, High Effort) (DeepSeek): kecerdasan 43.70, $109DeepSeek V4 Pro (Reasoning, Max Effort) (DeepSeek): kecerdasan 45.30, $176DeepSeek V4 Pro 0813 (Reasoning, Max Effort) (DeepSeek): kecerdasan 53.20, $605Gemini 2.5 Pro (Google): kecerdasan 25.90, $611Gemini 3.1 Flash-Lite (Google): kecerdasan 25.60, $93.8Gemini 3.1 Pro Preview (Google): kecerdasan 47.70, $812Gemini 3.5 Flash (high) (Google): kecerdasan 52.00, $1042Gemini 3.5 Flash-Lite (Google): kecerdasan 37.40, $153Gemini 3.6 Flash (high) (Google): kecerdasan 51.60, $412Gemma 4 26B A4B (Reasoning) (Google): kecerdasan 26.10, $53.9Gemma 4 31B (Non-reasoning) (Google): kecerdasan 22.30, $33.8GLM-4.6 (Reasoning) (Z AI): kecerdasan 29.30, $314GLM-4.7 (Reasoning) (Z AI): kecerdasan 34.50, $607GLM-5.1 (Reasoning) (Z AI): kecerdasan 41.00, $700GLM-5.2 (max) (Z AI): kecerdasan 52.60, $843GLM-5.3 (max) (Z AI): kecerdasan 59.50, $1239GPT-4.1 mini (OpenAI): kecerdasan 14.80, $64.2GPT-4.1 nano (OpenAI): kecerdasan 9.60, $36.7GPT-5 (high) (OpenAI): kecerdasan 35.30, $789GPT-5 mini (high) (OpenAI): kecerdasan 25.80, $130GPT-5.1 (high) (OpenAI): kecerdasan 37.50, $782GPT-5.4 (xhigh) (OpenAI): kecerdasan 53.10, $2183GPT-5.4 mini (xhigh) (OpenAI): kecerdasan 40.90, $1096GPT-5.4 nano (xhigh) (OpenAI): kecerdasan 39.70, $277GPT-5.5 (high) (OpenAI): kecerdasan 54.70, $1721GPT-5.5 (low) (OpenAI): kecerdasan 44.50, $374GPT-5.5 (medium) (OpenAI): kecerdasan 51.40, $939GPT-5.5 (Non-reasoning) (OpenAI): kecerdasan 35.80, $212GPT-5.5 (xhigh) (OpenAI): kecerdasan 56.30, $2778GPT-5.5 Instant (June 2026) (OpenAI): kecerdasan 29.20, $423GPT-5.6 Sol (low) (OpenAI): kecerdasan 50.70, $344GPT-5.6 Sol (max) (OpenAI): kecerdasan 60.90, $2823GPT-5.6 Sol (medium) (OpenAI): kecerdasan 55.60, $580GPT-5.6 Sol (Non-reasoning) (OpenAI): kecerdasan 41.90, $240GPT-5.6 Sol (xhigh) (OpenAI): kecerdasan 59.00, $1525GPT-5.6 Terra (high) (OpenAI): kecerdasan 50.10, $395GPT-5.6 Terra (low) (OpenAI): kecerdasan 41.30, $130GPT-5.6 Terra (max) (OpenAI): kecerdasan 56.60, $1390GPT-5.6 Terra (medium) (OpenAI): kecerdasan 46.80, $192GPT-5.6 Terra (Non-reasoning) (OpenAI): kecerdasan 34.60, $98.5GPT-5.6 Terra (xhigh) (OpenAI): kecerdasan 52.80, $590gpt-oss-120b (high) (OpenAI): kecerdasan 24.10, $95.0gpt-oss-120b (low) (OpenAI): kecerdasan 14.90, $24.2gpt-oss-20b (high) (OpenAI): kecerdasan 15.20, $32.7Grok 4.3 (high) (SpaceXAI): kecerdasan 37.90, $288Grok 4.3 (Non-reasoning) (SpaceXAI): kecerdasan 25.00, $201Grok 4.5 (high) (SpaceXAI): kecerdasan 55.80, $579Grok Build 0.1 0616 (SpaceXAI): kecerdasan 40.70, $377Hy3 (Tencent): kecerdasan 42.20, $93.2Inkling (xhigh) (Thinking Machines): kecerdasan 42.30, $698Inkling Small (Thinking Machines): kecerdasan 41.20, $193Kimi K2.5 (Reasoning) (Kimi): kecerdasan 36.00, $304Kimi K2.6 (Kimi): kecerdasan 45.10, $841Kimi K2.7 Code (Kimi): kecerdasan 43.00, $545Kimi K3 (low) (Kimi): kecerdasan 48.30, $283Kimi K3 (max) (Kimi): kecerdasan 59.70, $2425Ling 3.0 Flash (InclusionAI): kecerdasan 37.80, $72.8Llama 4 Maverick (Meta): kecerdasan 14.50, $35.9Llama 4 Scout (Meta): kecerdasan 10.30, $11.6LongCat 2.0 (LongCat): kecerdasan 34.00, $442Magistral Medium 1.2 (Mistral): kecerdasan 18.00, $920Magistral Small 1.2 (Mistral): kecerdasan 11.50, $301Mercury 2 (Inception): kecerdasan 21.90, $97.5MiMo-V2.5 (Xiaomi): kecerdasan 38.00, $25.2MiMo-V2.5-Pro (Xiaomi): kecerdasan 42.90, $98.9MiniMax-M2.7 (MiniMax): kecerdasan 38.90, $143MiniMax-M3 (MiniMax): kecerdasan 45.40, $205Ministral 3 14B (Mistral): kecerdasan 11.20, $119Ministral 3 3B (Mistral): kecerdasan 7.10, $91.6Ministral 3 8B (Mistral): kecerdasan 9.00, $114Mistral Large 3 (Mistral): kecerdasan 15.90, $71.1Mistral Medium 3.1 (Mistral): kecerdasan 14.70, $134Mistral Medium 3.5 (Mistral): kecerdasan 30.40, $883Mistral Small 3.1 (Mistral): kecerdasan 14.90, $29.8Mistral Small 3.2 (Mistral): kecerdasan 10.70, $129Mistral Small 4 (Reasoning) (Mistral): kecerdasan 19.70, $87.7Muse Glimmer (high) (Meta): kecerdasan 35.10, $105Muse Spark 1.1 (xhigh) (Meta): kecerdasan 53.20, $547Nemotron 3 Super 120B A12B (Reasoning) (NVIDIA): kecerdasan 25.70, $254Nemotron 3 Ultra 550B A55B (Reasoning) (NVIDIA): kecerdasan 38.30, $534Nemotron 3.5 Lightning (NVIDIA): kecerdasan 23.60, $71.5NVIDIA Nemotron 3 Nano 30B A3B (Reasoning) (NVIDIA): kecerdasan 14.50, $39.4Qwen3 235B A22B 2507 (Reasoning) (Alibaba): kecerdasan 19.90, $170Qwen3 30B A3B 2507 (Reasoning) (Alibaba): kecerdasan 14.60, $152Qwen3 Coder Next (Alibaba): kecerdasan 21.30, $250Qwen3 Next 80B A3B (Reasoning) (Alibaba): kecerdasan 16.90, $84.0Qwen3.5 122B A10B (Non-reasoning) (Alibaba): kecerdasan 28.20, $221Qwen3.5 122B A10B (Reasoning) (Alibaba): kecerdasan 32.80, $444Qwen3.5 35B A3B (Non-reasoning) (Alibaba): kecerdasan 24.30, $228Qwen3.5 397B A17B (Reasoning) (Alibaba): kecerdasan 34.30, $527Qwen3.5 9B (Reasoning) (Alibaba): kecerdasan 21.80, $240Qwen3.6 27B (Non-reasoning) (Alibaba): kecerdasan 31.30, $359Qwen3.6 27B (Reasoning) (Alibaba): kecerdasan 37.70, $667Qwen3.6 35B A3B (Non-reasoning) (Alibaba): kecerdasan 24.60, $530Qwen3.6 35B A3B (Reasoning) (Alibaba): kecerdasan 32.10, $505Qwen3.6 Plus (Alibaba): kecerdasan 40.50, $526Qwen3.7 Max (Alibaba): kecerdasan 46.70, $1064Qwen3.7 Plus (Alibaba): kecerdasan 39.40, $322Qwen3.8 2.4T A95B (Alibaba): kecerdasan 57.70, $1672Qwen3.8 Max (Alibaba): kecerdasan 58.10, $1741Ring-2.6-1T (InclusionAI): kecerdasan 31.70, $455Solar Pro 3 (Upstage): kecerdasan 14.50, $158Solar Pro 4 (Upstage): kecerdasan 41.60, $321Step 3.7 Flash (StepFun): kecerdasan 30.90, $320GPT-5.6 Luna (Non-reasoning) (OpenAI): kecerdasan 26.80, $10.2GPT-5.6 Luna (low) (OpenAI): kecerdasan 33.90, $13.7GPT-5.6 Luna (medium) (OpenAI): kecerdasan 38.90, $21.1GPT-5.6 Luna (high) (OpenAI): kecerdasan 47.00, $55.0GPT-5.6 Luna (xhigh) (OpenAI): kecerdasan 50.10, $95.1Gemini 3.7 Flash (low) (Google): kecerdasan 50.90, $165GPT-5.6 Luna (max) (OpenAI): kecerdasan 52.30, $172Gemini 3.7 Flash (medium) (Google): kecerdasan 53.40, $278Gemini 3.7 Flash (high) (Google): kecerdasan 56.00, $485Muse Spark 1.2 (xhigh) (Meta): kecerdasan 56.80, $639GPT-5.6 Sol (high) (OpenAI): kecerdasan 57.30, $955Grok 4.6 (high) (SpaceXAI): kecerdasan 60.90, $1068Claude Opus 5 (Adaptive Reasoning, High Effort) (Anthropic): kecerdasan 61.50, $1974Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) (Anthropic): kecerdasan 62.50, $2909Claude Opus 5 (Adaptive Reasoning, Max Effort) (Anthropic): kecerdasan 63.10, $3836Trinity Large Thinking (Arcee AI): kecerdasan 18.70, $255Claude 4.5 Haiku (Reasoning) (Anthropic): kecerdasan 29.90, $471Claude 4.5 Haiku (Reasoning)Trinity Large ThinkingGPT-5.6 Luna (Non-reasoning)Claude Opus 5 (max)
Claude 4.5 Haiku (Reasoning)Trinity Large ThinkingPareto frontier hari iniBatas Pareto model dengan bobot yang dapat diunduh (open-weights frontier)Semua model lainnya
Perbandingan langsung

Bandingkan sesuatu yang lain

Agen: Accept: text/markdown or ?format=md mengembalikan halaman ini sebagai Markdown. JSON: /api/v1/compare/claude-4-5-haiku-reasoning/trinity-large-thinking · Markdown: /api/v1/compare/claude-4-5-haiku-reasoning/trinity-large-thinking?format=md · llms.txt · keterampilan agen