Grok 4.20 price cut, DeepSeek V4 Pro price cut (18 Oct 2026)
Grok 4.20 got cheaper; DeepSeek V4 Pro got cheaper. 16 price changes, 812 new benchmark results.
BenchLeader refreshed 1576 models from 24 of 24 data feeds (benchmark publishers, aggregators and price lists). 16 price changes, 812 new or updated benchmark results across 24 data feeds.
Price changes
- DeepSeek V4.1 Flash (max) cost per task moved from 0.3 to 0.5 (83% up). (details)
- DeepSeek V4 Flash (max) input price moved from $0.11/M to $0.14/M (27% up). (details)
- DeepSeek V4 Flash (max) output price moved from $0.23/M to $0.28/M (22% up). (details)
- DeepSeek V4 Flash (max) cost per task moved from 0.1 to 0.2 (95% up). (details)
- Grok 4.20 (thinking) input price moved from $2.00/M to $1.25/M (38% down). (details)
- Grok 4.20 (thinking) output price moved from $6.00/M to $2.50/M (58% down). (details)
- Deepseek v4 Flash Vision (max) cost per task moved from 0.3 to 1 (230% up). (details)
- DeepSeek V4 Pro (high) input price moved from $0.66/M to $0.44/M (33% down). (details)
- DeepSeek V4 Pro (high) output price moved from $1.98/M to $0.87/M (56% down). (details)
- DeepSeek V4.1 Flash (no reasoning) cost per task moved from 0.1 to 0.2 (51% up). (details)
- Grok 4.20 (no reasoning) input price moved from $2.00/M to $1.25/M (38% down). (details)
- Grok 4.20 (no reasoning) output price moved from $6.00/M to $2.50/M (58% down). (details)
- GLM 4.6V (thinking) cached input price moved from $0.055/M to $0.050/M (9% down). (details)
- GLM 4.6V (no reasoning) cached input price moved from $0.055/M to $0.050/M (9% down). (details)
- GLM 4.6V cached input price moved from $0.055/M to $0.050/M (9% down). (details)
- DeepSeek V4.1 Flash 1017 input price moved from $0.15/M to $0.20/M (33% up). (details)
New benchmark results
- AutomationBench: 160 new results, including Claude Fable 5.1, GPT-5.6 Sol, Claude Fable 5.1… (board)
- GDP.pdf (AA): 159 new results, including Claude Fable 5.1, GPT-5.6 Sol, Claude Fable 5.1… (board)
- MLCR: 58 new results, including GPT-5.6 Sol, GPT-5.6 Sol, GPT-5.6 Sol… (board)
- EnterpriseOps-Gym: 32 new results, including GPT-5.6 Sol, GPT-5.5, GPT-5.6 Terra… (board)
- AA-Omniscience: accuracy: 29 new results, including Claude Fable 5, Claude Opus 5, Claude Opus 5… (board)
- AA-Omniscience: non-hallucination: 29 new results, including Claude Fable 5, Claude Opus 5, Claude Opus 5… (board)
- τ³-bench Banking: 22 new results, including Claude Fable 5, Claude Opus 5, GPT-5.6 Sol… (board)
- GDPval-AA v2.1: 21 new results, including Claude Fable 5, Claude Opus 5, Claude Opus 5… (board)
- Terminal-Bench-Science 0.1 (AA): 21 new results, including GPT-5.6 Sol, Claude Opus 5.5, GPT-6 Sol… (board)
- AA Intelligence Index v4.3.2: 17 new results, including GPT-6 Sol, Grok 4.6, GPT-6 Sol… (board)
- Humanity's Last Exam (AA): 17 new results, including GPT-6 Sol, Grok 4.6, GPT-6 Sol… (board)
- AA-LCR: 15 new results, including GPT-6 Sol, Grok 4.6, GPT-6 Sol… (board)
Score revisions
- Qwen3.8 Max (max) on SciCode (AA): 52.1% → 53.2%. (details)
- Qwen3.8 Max (max) on CritPt: 17.7% → 20%. (details)
- Qwen3.8 Max (max) on τ³-Banking (AA): 47.8% → 51.3%. (details)
- DeepSeek V4 Pro (max) on Terminal-Bench 4.0 (AA): 14.1% → 14.7%. (details)
- DeepSeek V4 Pro (max) on AA-Omniscience: non-hallucination: 5.2% → 5.9%. (details)
- DeepSeek V4 Pro (max) on GDP.pdf: 11.4% → 13.4%. (details)
- DeepSeek V4 Pro (max) on MLCR: 17.8% → 21.1%. (details)
- DeepSeek V4 Flash (max) on SciCode: 44.9% → 49.9%. (details)
- DeepSeek V4 Flash (max) on WeirdML: 45.6% → 63%. (details)
- DeepSeek V4 Flash (max) on LMCA: 35.9% → 41.7%. (details)
- DeepSeek V4 Flash (max) on DTBench: 86.4% → 90.9%. (details)
- DeepSeek V4 Flash (high) on WeirdML: 43.8% → 57.1%. (details)
- GPT-5.5 Instant on GPQA Diamond (AA): 82.3% → 84.7%. (details)
- GPT-5.5 Instant on Humanity's Last Exam (AA): 19.9% → 21.6%. (details)
- GPT-5.5 Instant on AA-Omniscience: 3.7 → 11.2. (details)
- GPT-5.5 Instant on CritPt: 0% → 2.6%. (details)
- GPT-5.5 Instant on AA-Omniscience: accuracy: 44.4% → 46.3%. (details)
- GPT-5.5 Instant on AA-Omniscience: non-hallucination: 26.8% → 34.7%. (details)
- Motif 3 on GPQA Diamond (AA): 83.4% → 86.9%. (details)
- Motif 3 on Humanity's Last Exam (AA): 37% → 40.4%. (details)
- …and 29 more.
Speed changes
- GPT-6 Astra time to first answer changed from 47.50 s to 88.82 s. (details)
- GPT-6 Astra response time changed from 56.96 s to 98.10 s. (details)
- Claude Opus 5.5 time to first answer changed from 34.99 s to 47.53 s. (details)
- Claude Fable 5.1 time to first answer changed from 7.87 s to 19.28 s. (details)
- Claude Fable 5.1 response time changed from 16.40 s to 27.81 s. (details)
- Claude Sonnet 5.5 time to first token changed from 0.98 s to 2.18 s. (details)
- Muse Spark 1.3 time to first token changed from 2.65 s to 3.63 s. (details)
- Claude Sonnet 5.5 time to first answer changed from 24.63 s to 35.54 s. (details)
- Claude Sonnet 5.5 response time changed from 29.02 s to 39.95 s. (details)
- GPT-5.6 Sol time to first answer changed from 9.07 s to 22.60 s. (details)
- GPT-5.6 Sol response time changed from 14.35 s to 28.33 s. (details)
- GPT-5.5 time to first answer changed from 37.05 s to 53.18 s. (details)
- GPT-5.5 response time changed from 42.12 s to 58.12 s. (details)
- GPT-5.5 time to first token changed from 9.41 s to 2.89 s. (details)
- GPT-6 Sol output speed changed from 78 tok/s to 113 tok/s. (details)
- …and 115 more.
Today's top five
- Claude Opus 5.5 (max) — 72.6
- Gemini 4 Argon (high) — 70.7
- Muse Wonder Full (max) — 70.6
- Claude Fable 5.1 (max) — 70.5
- GPT-6 Astra (max) — 70.4
More: New AI model releases this week, and the full LLM leaderboard 2026.
This digest is generated automatically from the day's data changes. Every line links to the page where you can check the numbers and their source.