Nanbeige4.3-33B added, GLM 5.3 FlashX Legal added (2 Nov)

BenchLeader refreshed 1681 models from 27 of 27 data feeds (benchmark publishers, aggregators and price lists). 2 new models, 2 price changes, 41 new or updated benchmark results across 27 data feeds.

New models

Nanbeige4.3-33B (Nanbeige) appeared with 4 benchmark results. (details)
GLM 5.3 FlashX Legal (Zhipu AI) is now listed at $0.59/M blended; no independent results yet. (details)

Price changes

DeepSeek V4.1 Flash 1017 (no reasoning) input price moved from $0.45/M to $0.30/M (33% down). (details)
DeepSeek V4.1 Flash 1017 (no reasoning) output price moved from $1.80/M to $1.20/M (33% down). (details)

New benchmark results

GameBench 2 by reasoning setting: 5 new results, including DeepSeek V4 Flash, Gemini 3.5 Flash, Gemini 3.5 Flash… (board)
FrontierCode 2.0 (AA): 5 new results, including DeepSeek V4.1 Flash 1017, Granite 4.2 8B, K2 Horizon 7B… (board)
Terminal-Bench-Science 0.1 (AA): 4 new results, including DeepSeek V4.1 Flash 1017, Granite 4.2 30B, K2 Horizon 7B… (board)
AA Intelligence Index v5.0: 2 new results, including DeepSeek V4.1 Flash 1017, Granite 4.2 8B (board)
CritPt v2 hard subset (AA): 2 new results, including DeepSeek V4.1 Flash 1017, K2 Horizon 7B (board)
AA Progress Index v1: 1 new result, including DeepSeek V4.1 Flash 1017 (board)
HLE-Diamond (AA): 1 new result, including DeepSeek V4.1 Flash 1017 (board)
MMMU-Pro: 1 new result, including DeepSeek V4.1 Flash 1017 (board)
AA-Omniscience v1.1: 1 new result, including DeepSeek V4.1 Flash 1017 (board)
Terminal-Bench 4.0 (AA): 1 new result, including DeepSeek V4.1 Flash 1017 (board)
AA-Omniscience v1.1: accuracy: 1 new result, including DeepSeek V4.1 Flash 1017 (board)
AA-Omniscience v1.1: non-hallucination: 1 new result, including DeepSeek V4.1 Flash 1017 (board)

Score revisions

Gemini 4 Carbon (high) on Vals Finance Index: 66.4% → 67.1%. (details)
Gemini 4 Carbon (high) on Tax Agent Bench: 48.9% → 51.8%. (details)
GPT-6.1 Terra (max) on Tax Agent Bench: 15.2% → 17.1%. (details)
GPT-5.6 Sol (medium) on GameBench 2 by reasoning setting: 67.2 → 68. (details)
Gemini 3.8 Flash (medium) on GameBench 2 by reasoning setting: 60.7 → 61.2. (details)
GPT-5 (medium) on GameBench 2 by reasoning setting: 37.8 → 38.6. (details)
GPT-5.4 Mini (no reasoning) on GameBench 2 by reasoning setting: 30.2 → 29.6. (details)
Claude Fable 5 (xhigh) on GameBench 2 by reasoning setting: 75.5 → 76.9. (details)
Mistral Large 4 OSS (high) on Vals Finance Index: 56.2% → 57%. (details)
Mistral Large 4 OSS (high) on Tax Agent Bench: 27.3% → 30.7%. (details)
Nemotron 3.5 Super VL 120B A12B (high) on Tax Agent Bench: 10.6% → 11.6%. (details)
Qwen3.8 Max (low) on GameBench 2 by reasoning setting: 50.2 → 50.7. (details)
Seed-2.0-Code (high) on GameBench 2 by reasoning setting: 42.3 → 43. (board)

Speed changes

GPT-6 Astra time to first answer changed from 270.03 s to 367.57 s. (details)
GPT-6 Astra response time changed from 276.42 s to 374.56 s. (details)
Claude Opus 5.5 time to first answer changed from 34.52 s to 47.50 s. (details)
GPT-5.5 time to first answer changed from 47.55 s to 68.13 s. (details)
GPT-5.5 response time changed from 54.01 s to 73.64 s. (details)
GPT-6.1 Sol time to first answer changed from 44.36 s to 69.58 s. (details)
GPT-6.1 Sol response time changed from 52.70 s to 77.10 s. (details)
GPT-5.4 Pro time to first token changed from 14.44 s to 8.18 s. (details)
GPT-5.3-Codex output speed changed from 79 tok/s to 114 tok/s. (details)
GPT-5.3-Codex time to first token changed from 1.78 s to 2.58 s. (details)
Claude Fable 5.1 time to first answer changed from 3.25 s to 12.88 s. (details)
Claude Fable 5.1 response time changed from 12.38 s to 21.56 s. (details)
Muse Spark 1.3 time to first token changed from 5.28 s to 2.48 s. (details)
GPT-5.6 Sol time to first answer changed from 3.39 s to 7.42 s. (details)
GPT-6.1 Terra time to first answer changed from 45.76 s to 68.83 s. (details)
…and 109 more.

Today's top five

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.