Add DeepSeek V4 Flash 0731 to live bench

[anthonypizzolato-abacus]

Add DeepSeek V4 Flash 0731 to live bench

Registers the DeepSeek-V4-Flash-0731 build (released 2026-07-31) in LanguageModelList.

Field	Value
Model name (display)	DeepSeek-V4-Flash-0731
Provider API ID (model_name)	deepseek-v4-flash
LMStyle	LMStyle.DeepSeekAPI (DeepSeek first-party API, https://api.deepseek.com)
Context window	1,000,000 tokens
Max output	384,000 tokens
Release/knowledge cutoff	2025-07-01
Link	https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731

Notes

• deepseek-v4-flash is the model ID DeepSeek's first-party API uses for the 0731 build (the preview and 0731 share this ID; the 0731 build is a post-training refinement of the same 284B/13B-active MoE architecture). DeepSeekRunner sends args.model as model to https://api.deepseek.com, so this ID resolves to the 0731 build.
• The 0731 build supports a reasoning_effort parameter (low/high/max) for thinking; default chat style DeepSeekAPI is used here, matching the existing deepseek-chat / deepseek-coder entries.
• Added adjacent to the existing DeepSeek API entries to keep the DeepSeek grouping.