The falling price of intelligence

One dot per model: its release date against how much measured intelligence $1 buys — Artificial Analysis Intelligence Index score divided by the model's cheapest blended price per 1M tokens, on a log scale. The climb is the story: each generation delivers more reasoning per dollar, one to two orders of magnitude in two years. Click a brand to isolate it; click a dot for its page.

131030100300Jul 20242025Jul 20252026Jul 2026index points per $ (log)GLM 4.5 Air (Novita) — 17 index pts/$ · score 16.7 at $0.98/1M · released 2025-07-28GLM 4.6 (Venice) — 10.7 index pts/$ · score 23.4 at $2.18/1M · released 2025-09-30GLM 4.7 (DeepInfra) — 16 index pts/$ · score 34.5 at $2.15/1M · released 2025-12-22GLM 4.7 Flash (Venice) — 50.7 index pts/$ · score 23.3 at $0.46/1M · released 2026-01-19GLM 5 (GMICloud) — 16.1 index pts/$ · score 40.5 at $2.52/1M · released 2026-02-12GLM 5.1 (Baidu) — 10.9 index pts/$ · score 41 at $3.764/1M · released 2026-04-07GLM 5.2 (Baidu) — 35.6 index pts/$ · score 52.6 at $1.479/1M · released 2026-06-13GLM 5.3 (Reka) — 12.8 index pts/$ · score 59.5 at $4.65/1M · released 2026-08-14GLM 5.3 Flash (Relace) — 186.2 index pts/$ · score 57.5 at $0.309/1M · released 2026-08-26GPT-4o — 0.9 index pts/$ · score 11.1 at $12.5/1M · released 2024-05-13GPT-4o mini — 8.9 index pts/$ · score 6.7 at $0.75/1M · released 2024-07-18GPT-4.1 mini — 7.4 index pts/$ · score 14.8 at $2/1M · released 2025-04-14gpt-oss-20b (AkashML) — 126.7 index pts/$ · score 15.2 at $0.12/1M · released 2025-08-05gpt-oss-120b (AkashML) — 120.5 index pts/$ · score 24.1 at $0.2/1M · released 2025-08-05GPT-5.6 Sol (OpenAI) — 9.8 index pts/$ · score 58.9 at $6/1M · released 2026-07-09GPT-5.6 Terra (OpenAI) — 7.9 index pts/$ · score 55 at $7/1M · released 2026-07-09GPT-5.6 Luna (OpenAI) — 73.1 index pts/$ · score 51.2 at $0.7/1M · released 2026-07-09Qwen3.5 397B A17B (Alibaba) — 12.6 index pts/$ · score 34.3 at $2.73/1M · released 2026-02-15Qwen3.5-35B-A3B (DeepInfra) — 26.2 index pts/$ · score 29.9 at $1.14/1M · released 2026-02-23Qwen3.5-27B (Alibaba) — 19.7 index pts/$ · score 34.6 at $1.755/1M · released 2026-02-23Qwen3.5-122B-A10B (SiliconFlow) — 14.1 index pts/$ · score 32.9 at $2.34/1M · released 2026-02-23Qwen3.6 35B A3B (AkashML) — 32.1 index pts/$ · score 32.1 at $1/1M · released 2026-04-17Qwen3.6 27B (Chutes) — 16.4 index pts/$ · score 37.7 at $2.3/1M · released 2026-04-22Qwen3.8 27B (Parasail) — 21.3 index pts/$ · score 52 at $2.44/1M · released 2026-08-14DeepSeek V3 (DeepInfra) — 11.7 index pts/$ · score 14.2 at $1.21/1M · released 2024-12-26DeepSeek R1 (DeepInfra) — 7.7 index pts/$ · score 20.4 at $2.65/1M · released 2025-05-28DeepSeek V3.1 (DeepInfra) — 17.8 index pts/$ · score 21.4 at $1.2/1M · released 2025-08-21DeepSeek V3.2 (GMICloud) — 48.4 index pts/$ · score 25.1 at $0.518/1M · released 2025-12-01DeepSeek V4 Flash 0731 (Baidu) — 345.3 index pts/$ · score 51.8 at $0.15/1M · released 2026-04-24Gemma 3 27B (DeepInfra) — 30.8 index pts/$ · score 7.4 at $0.24/1M · released 2025-03-12Gemma 4 31B (CoreWeave) — 67.5 index pts/$ · score 29.7 at $0.44/1M · released 2026-04-02Gemma 4 26B A4B (Cloudflare) — 65.2 index pts/$ · score 26.1 at $0.4/1M · released 2026-04-02Gemini 3.6 Flash — 11.5 index pts/$ · score 51.6 at $4.5/1M · released 2026-07-21Gemini 3.8 Flash (Google) — 13 index pts/$ · score 58.7 at $4.5/1M · released 2026-09-02Mistral Medium 3 (Mistral) — 5.2 index pts/$ · score 12.5 at $2.4/1M · released 2025-05-07Ministral 3 14B 2512 (Mistral) — 28 index pts/$ · score 11.2 at $0.4/1M · released 2025-12-02Ministral 3 8B 2512 (Mistral) — 30 index pts/$ · score 9 at $0.3/1M · released 2025-12-02Ministral 3 3B 2512 (Mistral) — 35.5 index pts/$ · score 7.1 at $0.2/1M · released 2025-12-02Mistral Medium 3.5 (Mistral) — 3.4 index pts/$ · score 30.4 at $9/1M · released 2026-04-30MiniMax M1 (Minimax) — 6.9 index pts/$ · score 17.9 at $2.6/1M · released 2025-06-17MiniMax M2.5 (Venice) — 28.3 index pts/$ · score 34.5 at $1.22/1M · released 2026-02-12MiniMax M2.7 (Novita) — 28.8 index pts/$ · score 38.9 at $1.35/1M · released 2026-03-18MiniMax M3 (CoreWeave) — 38.2 index pts/$ · score 45.4 at $1.19/1M · released 2026-06-01Kimi K2.5 (SiliconFlow) — 13.3 index pts/$ · score 36 at $2.7/1M · released 2026-01-27Kimi K2.6 (Baidu) — 15.8 index pts/$ · score 45.1 at $2.856/1M · released 2026-04-21Kimi K2.7 Code (DeepInfra) — 10.5 index pts/$ · score 43 at $4.08/1M · released 2026-06-12Kimi K3 (DeepInfra) — 3.5 index pts/$ · score 59.7 at $17.1/1M · released 2026-07-16Claude Fable 5 (Anthropic) — 1 index pts/$ · score 62.1 at $60/1M · released 2026-06-09Claude Sonnet 5 (Anthropic) — 4.6 index pts/$ · score 55.3 at $12/1M · released 2026-06-30Claude Opus 5 (Anthropic) — 2.1 index pts/$ · score 63 at $30/1M · released 2026-07-24Nemotron 3 Super (DeepInfra) — 53 index pts/$ · score 25.7 at $0.485/1M · released 2026-03-11Nemotron 3 Ultra (DeepInfra) — 14.2 index pts/$ · score 38.3 at $2.7/1M · released 2026-06-04Nemotron 3.5 Lightning (DeepInfra) — 84.3 index pts/$ · score 23.6 at $0.28/1M · released 2026-08-11Llama 4 Scout (DeepInfra) — 25.8 index pts/$ · score 10.3 at $0.4/1M · released 2025-04-05Llama 4 Maverick (DigitalOcean) — 16.2 index pts/$ · score 14.5 at $0.896/1M · released 2025-04-05Granite 4.2 8B (CoreWeave) — 78.4 index pts/$ · score 19.6 at $0.25/1M · released 2026-08-31Grok 4.6 (xAI) — 7.6 index pts/$ · score 60.9 at $8/1M · released 2026-08-12

57 models with a benchmark score, verified 2026-09-05. Intelligence =Artificial Analysis Intelligence Index(0–100); price = cheapest live blended offer ($0 free tiers excluded). See also the raw price trend and thecurrent intelligence-per-dollar ranking ·methodology.

FAQ

How fast is the price of intelligence falling?

Fast. On this chart the best intelligence-per-dollar offer has climbed to 345.3 index points per dollar (DeepSeek V4 Flash 0731 (Baidu)), while flagships from a year or two earlier still sit one to two orders of magnitude lower. Each model generation buys more measured reasoning for the same dollar.

What does 'intelligence per dollar' mean?

Each model's Artificial Analysis Intelligence Index score (an aggregate 0–100 reasoning and knowledge benchmark) divided by its cheapest blended API price per 1M tokens. It answers: how much benchmark performance does $1 of tokens buy?

Why do the newest expensive flagships sit low on this chart?

Because peak capability carries a price premium. A flagship can score highest in raw intelligence yet buy the least intelligence per dollar. The models that maximize this metric are usually small, recent, and aggressively priced.