LLM Cost Frontier: frontier advanceshttps://llm-cost-frontier.catalystneuro.com/feed.xml2026-09-09T06:00:00ZCatalystNeuroEach entry is a date on which a model became the cheapest way to reach some level of the Artificial Analysis Intelligence Index, through a release or a price change.2026-09-03: GPT-6 Astra (xhigh) (new model, index 61.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-09-03/gpt-6-astra-xhigh2026-09-03T00:00:00ZGPT-6 Astra (xhigh): now the cheapest way to reach index 60.9 to 61.0 at $1.20 per task, taking it from Claude Opus 5. Proprietary.2026-09-02: Muse Spark 1.3 (xhigh) (new model, index 60.8)https://llm-cost-frontier.catalystneuro.com/advance/2026-09-02/muse-spark-1-3-xhigh2026-09-02T00:00:00ZMuse Spark 1.3 (xhigh): now the cheapest way to reach index 57.5 to 60.8 at $0.55 per task, taking it from Grok 4.6, Kimi K3, GLM-5.3, and GPT-5.6 Sol; Kimi K3, GLM-5.3, and GPT-5.6 Sol left the frontier. New cost record for index ≥ 60. Proprietary.2026-09-01: Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback) (new model, index 65.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-09-01/claude-fable-5-12026-09-01T00:00:00ZClaude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback): pushed the intelligence ceiling from 63.1 to 65.7, at $3.69 per task. Proprietary.2026-09-01: Claude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback) (new model, index 64.8)https://llm-cost-frontier.catalystneuro.com/advance/2026-09-01/claude-fable-5-1-xhigh2026-09-01T00:00:00ZClaude Fable 5.1 (Adaptive Reasoning, Xhigh Effort, Default Fallback): pushed the intelligence ceiling from 63.1 to 64.8, at $2.65 per task. Proprietary.2026-09-01: Claude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback) (new model, index 62.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-09-01/claude-fable-5-1-high2026-09-01T00:00:00ZClaude Fable 5.1 (Adaptive Reasoning, High Effort, Default Fallback): now the cheapest way to reach index 61.5 to 62.5 at $1.43 per task, taking it from Claude Opus 5. Proprietary.2026-08-28: GPT-5.6 Sol (xhigh) (price change, index 59.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-28/gpt-5-6-sol-xhigh2026-08-28T00:00:00ZGPT-5.6 Sol (xhigh): price moved from $0.671 to $0.63 per task; now the cheapest way to reach index 57.5 to 59.0. Proprietary.2026-08-26: GLM-5.3-Flash (new model, index 57.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-26/glm-5-3-flash2026-08-26T00:00:00ZGLM-5.3-Flash: now the cheapest way to reach index 52.3 to 57.5 at $0.087 per task, taking it from GPT-5.6 Sol, Muse Spark 1.2, Grok 4.5, Gemini 3.7 Flash, and DeepSeek V4 Pro 0813; Muse Spark 1.2, Grok 4.5, Gemini 3.7 Flash, and DeepSeek V4 Pro 0813 left the frontier. Open weights.2026-08-26: Agnes 2.5 Pro Beta (new model, index 49.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-26/agnes-2-5-pro-beta2026-08-26T00:00:00ZAgnes 2.5 Pro Beta: now the cheapest way to reach index 47.0 to 49.1 at $0.027 per task, taking it from GPT-5.6 Luna. Proprietary.2026-08-25: GPT-5.6 Sol (xhigh) (price change, index 59.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-25/gpt-5-6-sol-xhigh2026-08-25T00:00:00ZGPT-5.6 Sol (xhigh): price moved from $0.807 to $0.64 per task; now the cheapest way to reach index 57.3 to 59.0, taking it from Grok 4.6. Proprietary.2026-08-25: GPT-5.6 Sol (high) (price change, index 57.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-25/gpt-5-6-sol-high2026-08-25T00:00:00ZGPT-5.6 Sol (high): price moved from $0.548 to $0.43 per task; now the cheapest way to reach index 56.8 to 57.3. Proprietary.2026-08-25: GPT-5.6 Sol (medium) (price change, index 55.6)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-25/gpt-5-6-sol-medium2026-08-25T00:00:00ZGPT-5.6 Sol (medium): price moved from $0.372 to $0.30 per task; now the cheapest way to reach index 53.4 to 55.6, taking it from Grok 4.5. Proprietary.2026-08-25: Granite 4.2 3B (new model, index 14.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-25/granite-4-2-3b2026-08-25T00:00:00ZGranite 4.2 3B: now the cheapest way to reach index 0.0 to 14.3 at $0.0047 per task, taking it from GPT-5.6 Luna. Open weights.2026-08-18: GLM-5.3 (max) (new model, index 59.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-18/glm-5-32026-08-18T00:00:00ZGLM-5.3 (max): now the cheapest way to reach index 59.0 to 59.5 at $0.68 per task, taking it from Grok 4.6. Open weights.2026-08-13: Gemini 3.7 Flash (medium) (new model, index 53.4)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-13/gemini-3-7-flash-medium2026-08-13T00:00:00ZGemini 3.7 Flash (medium): now the cheapest way to reach index 53.2 to 53.4 at $0.26 per task, taking it from Grok 4.5. Proprietary.2026-08-13: DeepSeek V4 Pro 0813 (Reasoning, Max Effort) (new model, index 53.2)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-13/deepseek-v4-pro2026-08-13T00:00:00ZDeepSeek V4 Pro 0813 (Reasoning, Max Effort): now the cheapest way to reach index 52.3 to 53.2 at $0.25 per task, taking it from Muse Spark 1.1, which left the frontier. Open weights.2026-08-12: Grok 4.6 (high) (new model, index 60.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-12/grok-4-62026-08-12T00:00:00ZGrok 4.6 (high): now the cheapest way to reach index 59.0 to 60.9 at $0.84 per task, taking it from Claude Opus 5 and Kimi K3; Kimi K3 left the frontier. New cost record for index ≥ 60. Proprietary.2026-08-12: Grok 4.6 (medium) (new model, index 59.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-12/grok-4-6-medium2026-08-12T00:00:00ZGrok 4.6 (medium): now the cheapest way to reach index 57.3 to 59.0 at $0.67 per task, taking it from GPT-5.6 Sol and Claude Opus 5. Proprietary.2026-08-05: Muse Spark 1.2 (xhigh) (new model, index 56.8)https://llm-cost-frontier.catalystneuro.com/advance/2026-08-05/muse-spark-1-22026-08-05T00:00:00ZMuse Spark 1.2 (xhigh): now the cheapest way to reach index 55.8 to 56.8 at $0.40 per task, taking it from GPT-5.6 Sol and GPT-5.6 Terra; GPT-5.6 Terra left the frontier. Proprietary.2026-07-30: GPT-5.6 Terra (max) (price change, index 56.6)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-terra2026-07-30T00:00:00ZGPT-5.6 Terra (max): price moved from $0.635 to $0.51 per task; now the cheapest way to reach index 55.8 to 56.6, taking it from GPT-5.6 Sol. Proprietary.2026-07-30: GPT-5.6 Luna (max) (price change, index 52.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-luna2026-07-30T00:00:00ZGPT-5.6 Luna (max): price moved from $0.236 to $0.047 per task; now the cheapest way to reach index 50.1 to 52.3, taking it from GPT-5.6 Sol. Proprietary.2026-07-30: GPT-5.6 Luna (xhigh) (price change, index 50.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-luna-xhigh2026-07-30T00:00:00ZGPT-5.6 Luna (xhigh): price moved from $0.158 to $0.032 per task; now the cheapest way to reach index 47.0 to 50.1. New cost record for index ≥ 50. Proprietary.2026-07-30: GPT-5.6 Luna (high) (price change, index 47.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-luna-high2026-07-30T00:00:00ZGPT-5.6 Luna (high): price moved from $0.108 to $0.022 per task; now the cheapest way to reach index 38.9 to 47.0, taking it from DeepSeek V4 Pro and MiMo-V2.5-Pro, both of which left the frontier. New cost record for index ≥ 40. Proprietary.2026-07-30: GPT-5.6 Luna (medium) (price change, index 38.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-luna-medium2026-07-30T00:00:00ZGPT-5.6 Luna (medium): price moved from $0.056 to $0.011 per task; now the cheapest way to reach index 38.0 to 38.9, taking it from MiMo-V2.5-Pro. Proprietary.2026-07-30: GPT-5.6 Luna (low) (price change, index 33.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-30/gpt-5-6-luna-low2026-07-30T00:00:00ZGPT-5.6 Luna (low): price moved from $0.044 to $0.0088 per task; now the cheapest way to reach index 0.0 to 33.9, taking it from MiMo-V2.5. New cost record for index ≥ 30. Proprietary.2026-07-24: Claude Opus 5 (Adaptive Reasoning, Max Effort) (new model, index 63.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-24/claude-opus-52026-07-24T00:00:00ZClaude Opus 5 (Adaptive Reasoning, Max Effort): pushed the intelligence ceiling from 62.1 to 63.1, at $2.34 per task. Proprietary.2026-07-24: Claude Opus 5 (Adaptive Reasoning, Xhigh Effort) (new model, index 62.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-24/claude-opus-5-xhigh2026-07-24T00:00:00ZClaude Opus 5 (Adaptive Reasoning, Xhigh Effort): pushed the intelligence ceiling from 62.1 to 62.5, at $1.80 per task, taking it from Claude Fable 5, which left the frontier. Proprietary.2026-07-24: Claude Opus 5 (Adaptive Reasoning, High Effort) (new model, index 61.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-24/claude-opus-5-high2026-07-24T00:00:00ZClaude Opus 5 (Adaptive Reasoning, High Effort): now the cheapest way to reach index 59.7 to 61.5 at $1.23 per task, taking it from Claude Fable 5 and GPT-5.6 Sol. New cost record for index ≥ 60. Proprietary.2026-07-24: Claude Opus 5 (Adaptive Reasoning, Medium Effort) (new model, index 58.6)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-24/claude-opus-5-medium2026-07-24T00:00:00ZClaude Opus 5 (Adaptive Reasoning, Medium Effort): now the cheapest way to reach index 57.3 to 58.6 at $0.72 per task, taking it from GPT-5.6 Sol. Proprietary.2026-07-16: Kimi K3 (max) (new model, index 59.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-16/kimi-k32026-07-16T00:00:00ZKimi K3 (max): now the cheapest way to reach index 59.0 to 59.7 at $0.84 per task, taking it from GPT-5.6 Sol. Open weights.2026-07-09: GPT-5.6 Sol (max) (new model, index 60.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-sol2026-07-09T00:00:00ZGPT-5.6 Sol (max): now the cheapest way to reach index 59.0 to 60.9 at $1.23 per task, taking it from Claude Fable 5. New cost record for index ≥ 60. Proprietary.2026-07-09: GPT-5.6 Sol (xhigh) (new model, index 59.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-sol-xhigh2026-07-09T00:00:00ZGPT-5.6 Sol (xhigh): now the cheapest way to reach index 57.3 to 59.0 at $0.81 per task, taking it from Claude Fable 5. Proprietary.2026-07-09: GPT-5.6 Sol (high) (new model, index 57.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-sol-high2026-07-09T00:00:00ZGPT-5.6 Sol (high): now the cheapest way to reach index 55.8 to 57.3 at $0.55 per task, taking it from Claude Opus 4.8 and GPT-5.5, both of which left the frontier. Proprietary.2026-07-09: Muse Spark 1.1 (xhigh) (new model, index 53.2)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/muse-spark-1-12026-07-09T00:00:00ZMuse Spark 1.1 (xhigh): now the cheapest way to reach index 52.3 to 53.2 at $0.29 per task, taking it from Grok 4.5. Proprietary.2026-07-09: GPT-5.6 Luna (max) (new model, index 52.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-luna2026-07-09T00:00:00ZGPT-5.6 Luna (max): now the cheapest way to reach index 50.7 to 52.3 at $0.24 per task, taking it from Grok 4.5. Proprietary.2026-07-09: GPT-5.6 Sol (low) (new model, index 50.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-sol-low2026-07-09T00:00:00ZGPT-5.6 Sol (low): now the cheapest way to reach index 50.1 to 50.7 at $0.23 per task, taking it from Grok 4.5. Proprietary.2026-07-09: GPT-5.6 Luna (xhigh) (new model, index 50.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-luna-xhigh2026-07-09T00:00:00ZGPT-5.6 Luna (xhigh): now the cheapest way to reach index 47.0 to 50.1 at $0.16 per task, taking it from Grok 4.5 and Gemini 3.1 Pro Preview; Gemini 3.1 Pro Preview left the frontier. New cost record for index ≥ 50. Proprietary.2026-07-09: GPT-5.6 Luna (high) (new model, index 47.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-09/gpt-5-6-luna-high2026-07-09T00:00:00ZGPT-5.6 Luna (high): now the cheapest way to reach index 45.3 to 47.0 at $0.11 per task, taking it from Gemini 3.1 Pro Preview and MiniMax-M3; MiniMax-M3 left the frontier. Proprietary.2026-07-08: Grok 4.5 (high) (new model, index 55.8)https://llm-cost-frontier.catalystneuro.com/advance/2026-07-08/grok-4-52026-07-08T00:00:00ZGrok 4.5 (high): now the cheapest way to reach index 47.7 to 55.8 at $0.36 per task, taking it from GPT-5.5 and GLM-5.2; GLM-5.2 left the frontier. New cost record for index ≥ 50. Proprietary.2026-06-16: GLM-5.2 (max) (new model, index 52.6)https://llm-cost-frontier.catalystneuro.com/advance/2026-06-16/glm-5-22026-06-16T00:00:00ZGLM-5.2 (max): now the cheapest way to reach index 47.7 to 52.6 at $0.44 per task, taking it from GPT-5.5 and Gemini 3.5 Flash; Gemini 3.5 Flash left the frontier. New cost record for index ≥ 50. Open weights.2026-06-09: Claude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback) (new model, index 62.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-06-09/claude-fable-52026-06-09T00:00:00ZClaude Fable 5 (Adaptive Reasoning, Max Effort, Opus 4.8 Fallback): pushed the intelligence ceiling from 57.3 to 62.1, at $3.14 per task. New cost record for index ≥ 60. Proprietary.2026-06-01: MiniMax-M3 (new model, index 45.4)https://llm-cost-frontier.catalystneuro.com/advance/2026-06-01/minimax-m32026-06-01T00:00:00ZMiniMax-M3: now the cheapest way to reach index 45.3 to 45.4 at $0.14 per task, taking it from Gemini 3.1 Pro Preview. Open weights.2026-05-28: Claude Opus 4.8 (Adaptive Reasoning, Max Effort) (new model, index 57.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-05-28/claude-opus-4-82026-05-28T00:00:00ZClaude Opus 4.8 (Adaptive Reasoning, Max Effort): pushed the intelligence ceiling from 56.3 to 57.3, at $2.03 per task. Proprietary.2026-05-19: Gemini 3.5 Flash (high) (new model, index 52.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-05-19/gemini-3-5-flash2026-05-19T00:00:00ZGemini 3.5 Flash (high): now the cheapest way to reach index 51.4 to 52.0 at $0.69 per task, taking it from GPT-5.5. Proprietary.2026-04-24: DeepSeek V4 Pro (Reasoning, Max Effort) (new model, index 45.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-24/deepseek-v4-pro-04242026-04-24T00:00:00ZDeepSeek V4 Pro (Reasoning, Max Effort): now the cheapest way to reach index 43.7 to 45.3 at $0.047 per task, taking it from Gemini 3.1 Pro Preview and GPT-5.5. Open weights.2026-04-24: DeepSeek V4 Pro (Reasoning, High Effort) (new model, index 43.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-24/deepseek-v4-pro-0424-high2026-04-24T00:00:00ZDeepSeek V4 Pro (Reasoning, High Effort): now the cheapest way to reach index 42.9 to 43.7 at $0.043 per task, taking it from GPT-5.5. Open weights.2026-04-23: GPT-5.5 (xhigh) (new model, index 56.3)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-23/gpt-5-52026-04-23T00:00:00ZGPT-5.5 (xhigh): pushed the intelligence ceiling from 55.0 to 56.3, at $1.17 per task, taking it from Claude Opus 4.7, which left the frontier. Proprietary.2026-04-23: GPT-5.5 (high) (new model, index 54.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-23/gpt-5-5-high2026-04-23T00:00:00ZGPT-5.5 (high): now the cheapest way to reach index 51.4 to 54.7 at $0.80 per task, taking it from Claude Opus 4.7 and GPT-5.4; GPT-5.4 left the frontier. Proprietary.2026-04-23: GPT-5.5 (medium) (new model, index 51.4)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-23/gpt-5-5-medium2026-04-23T00:00:00ZGPT-5.5 (medium): now the cheapest way to reach index 47.7 to 51.4 at $0.50 per task, taking it from GPT-5.4. New cost record for index ≥ 50. Proprietary.2026-04-23: GPT-5.5 (low) (new model, index 44.5)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-23/gpt-5-5-low2026-04-23T00:00:00ZGPT-5.5 (low): now the cheapest way to reach index 42.9 to 44.5 at $0.26 per task, taking it from Gemini 3.1 Pro Preview. Proprietary.2026-04-22: MiMo-V2.5-Pro (new model, index 42.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-22/mimo-v2-5-pro2026-04-22T00:00:00ZMiMo-V2.5-Pro: now the cheapest way to reach index 38.0 to 42.9 at $0.034 per task, taking it from Gemini 3.1 Pro Preview, GLM-5.1, GPT-5.4 nano, and MiniMax-M2.7; GLM-5.1, GPT-5.4 nano, and MiniMax-M2.7 left the frontier. New cost record for index ≥ 40. Open weights.2026-04-22: MiMo-V2.5 (new model, index 38.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-22/mimo-v2-5-04242026-04-22T00:00:00ZMiMo-V2.5: now the cheapest way to reach index 0.0 to 38.0 at $0.010 per task, taking it from MiniMax-M2.7, Gemma 4 26B A4B, GPT-5 mini, gpt-oss-20b, gpt-oss-120b, and Llama 4 Scout; Gemma 4 26B A4B, GPT-5 mini, gpt-oss-20b, gpt-oss-120b, and Llama 4 Scout left the frontier. New cost record for index ≥ 30. Open weights.2026-04-16: Claude Opus 4.7 (Adaptive Reasoning, Max Effort) (new model, index 55.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-16/claude-opus-4-72026-04-16T00:00:00ZClaude Opus 4.7 (Adaptive Reasoning, Max Effort): pushed the intelligence ceiling from 53.1 to 55.0, at $2.23 per task. Proprietary.2026-04-07: GLM-5.1 (Reasoning) (new model, index 41.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-07/glm-5-12026-04-07T00:00:00ZGLM-5.1 (Reasoning): now the cheapest way to reach index 39.7 to 41.0 at $0.30 per task, taking it from Gemini 3.1 Pro Preview. New cost record for index ≥ 40. Open weights.2026-04-02: Gemma 4 26B A4B (Reasoning) (new model, index 26.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-04-02/gemma-4-26b-a4b2026-04-02T00:00:00ZGemma 4 26B A4B (Reasoning): now the cheapest way to reach index 25.8 to 26.1 at $0.039 per task, taking it from MiniMax-M2.7. Open weights.2026-03-18: MiniMax-M2.7 (new model, index 38.9)https://llm-cost-frontier.catalystneuro.com/advance/2026-03-18/minimax-m2-72026-03-18T00:00:00ZMiniMax-M2.7: now the cheapest way to reach index 25.8 to 38.9 at $0.078 per task, taking it from GPT-5.4 nano and Kimi K2.5; Kimi K2.5 left the frontier. New cost record for index ≥ 30. Open weights.2026-03-17: GPT-5.4 nano (xhigh) (new model, index 39.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-03-17/gpt-5-4-nano2026-03-17T00:00:00ZGPT-5.4 nano (xhigh): now the cheapest way to reach index 36.0 to 39.7 at $0.15 per task, taking it from Gemini 3.1 Pro Preview and GPT-5.1; GPT-5.1 left the frontier. Proprietary.2026-03-05: GPT-5.4 (xhigh) (new model, index 53.1)https://llm-cost-frontier.catalystneuro.com/advance/2026-03-05/gpt-5-42026-03-05T00:00:00ZGPT-5.4 (xhigh): pushed the intelligence ceiling from 48.4 to 53.1, at $1.10 per task, taking it from Claude Sonnet 4.6, which left the frontier. New cost record for index ≥ 50. Proprietary.2026-02-19: Gemini 3.1 Pro Preview (new model, index 47.7)https://llm-cost-frontier.catalystneuro.com/advance/2026-02-19/gemini-3-1-pro-preview2026-02-19T00:00:00ZGemini 3.1 Pro Preview: now the cheapest way to reach index 37.5 to 47.7 at $0.33 per task, taking it from Claude Sonnet 4.6. New cost record for index ≥ 40. Proprietary.2026-02-17: Claude Sonnet 4.6 (Adaptive Reasoning, Max Effort) (new model, index 48.4)https://llm-cost-frontier.catalystneuro.com/advance/2026-02-17/claude-sonnet-4-6-adaptive2026-02-17T00:00:00ZClaude Sonnet 4.6 (Adaptive Reasoning, Max Effort): pushed the intelligence ceiling from 37.5 to 48.4, at $1.22 per task. New cost record for index ≥ 40. Proprietary.2026-01-27: Kimi K2.5 (Reasoning) (new model, index 36.0)https://llm-cost-frontier.catalystneuro.com/advance/2026-01-27/kimi-k2-52026-01-27T00:00:00ZKimi K2.5 (Reasoning): now the cheapest way to reach index 25.8 to 36.0 at $0.096 per task, taking it from GPT-5.1, GPT-5, Claude 4.5 Haiku, and Gemini 2.5 Pro; GPT-5, Claude 4.5 Haiku, and Gemini 2.5 Pro left the frontier. New cost record for index ≥ 30. Open weights.