Changelog

A record of feature, data, and experience improvements released to Leaderboard.cn users.

v0.1.662026-10-31

Mistral Large 4 weights now open

Mistral Large 4 full weights were released on 2026-10-29 under the Mistral Ecosystem Open-Weight License. The model page now shows the open-weight status, license, and Hugging Face / ModelScope links. Previously added API-preview benchmark results are unaffected.
Added the data source "Mistral Large 4 Hugging Face Model Card" with the three figures unique to the model card: Artificial Analysis Intelligence Index v5 (26), Coding Agent Index (48%), and Lakera B3 (93.3%).

v0.1.652026-10-29

GPT-6.1 Terra added

Added GPT-6.1 Terra with model specifications, Standard and Fast API pricing, and 18 official evaluation results.

v0.1.642026-10-24

Kimi K3.1 Pro

Added Kimi K3.1 Pro with model details, API pricing, and 50 official evaluation results.
Fixed display issues on model detail pages.

v0.1.632026-10-20

MiniMax M3.1 Flash

Added MiniMax M3.1 Flash with a 1M-token context window, multimodal input, reasoning capabilities, and API pricing. Model weights are marked as not yet available.
Added 15 official results with source evidence, preserving separate ALE full-set, OSWorld 0808 Full, and corrected NL2Repo evaluations. CyberGym uses the precise table value of 82.2.

v0.1.622026-10-19

Arena leaderboard results retired

Rankings no longer use Arena leaderboard results, including Arena scores quoted in provider reports. The corresponding result and data-source pages have been removed.
Model profiles and results from other sources are preserved. Independent benchmarks such as WebArena, MathArena, and Arena-Hard are unaffected.

v0.1.612026-10-19

Six new models and official benchmark results

Added DeepSeek V4.1 Flash 1017, Muse Wonder, Muse Spark Open, Nemotron 3.5 Super VL, Nemotron 3.5 Bolt 74B-A9B, and Mistral Large 4, including specifications, published API pricing, and six official source reports.
Added 110 benchmark results while preserving checkpoint, precision, test-subset, and offline-evaluation distinctions. Ambiguous results retain their source values without being scored. Mistral Large 4 is marked as an API preview with weights not yet released.
Fixed dollar-price text in static model summaries so prices remain intact.

v0.1.602026-10-08

Claude Haiku 5.5

Added Claude Haiku 5.5 with 16 benchmark metrics and same-source model comparisons, totaling 52 results. Results preserve reasoning effort, tool use, the OSWorld offline subset, and length adjustments. API pricing includes separate tiers for prompts up to and over 100K tokens.

v0.1.592026-10-01

Gemini 4 Argon

Added Gemini 4 Argon with 19 metrics across 18 benchmarks and same-source model comparisons, totaling 73 results. Results preserve evaluation versions, subsets, tools, and run conditions. Model details note limited Fairwind access, the 1M output limit, and announced introductory pricing.

v0.1.582026-10-01

GPT-6.1 Sol

Added GPT-6.1 Sol with 22 benchmark metrics from its release and system card, plus same-source GPT-6 Sol and GPT-6 Astra comparisons for 66 results in total. Results preserve actual reasoning efforts, scores and costs across efforts, length adjustments, and evaluation conditions.

v0.1.572026-09-29

Claude Sonnet 5.5

Added Claude Sonnet 5.5 with specifications, API pricing, and results across 18 official benchmarks, including comparisons with models in the same report and notes on reasoning effort, tool use, length adjustment, and fallback settings.

Home