fix: DeepSeek V4 Flash outdated

models/deepseek/deepseek-v4-flash.toml

@@ -1,8 +1,12 @@
+# DeepSeek-V4-Flash-0731 official API (public beta): same architecture as preview, re-post-trained.
+# https://api-docs.deepseek.com/updates (2026-07-31)
+# https://api-docs.deepseek.com/quick_start/pricing/
+# https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731
 name = "DeepSeek V4 Flash"
 description = "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work"
 family = "deepseek-flash"
 release_date = "2026-04-24"
-last_updated = "2026-04-24"
+last_updated = "2026-07-31"
 attachment = false
 reasoning = true
 temperature = true
@@ -21,10 +25,29 @@ output = ["text"]
 
 [[weights]]
 label = "Hugging Face"
-url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"
+url = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
 
 [[benchmarks]]
-name = "SWE-Bench Verified"
-score = 79
-metric = "resolved"
-source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash"
+name = "Terminal-Bench"
+score = 82.7
+metric = "pass@1"
+version = "2.1"
+source = "https://api-docs.deepseek.com/updates"
+
+[[benchmarks]]
+name = "Toolathlon"
+score = 70.3
+metric = "verified"
+source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+
+[[benchmarks]]
+name = "NL2Repo"
+score = 54.2
+metric = "score"
+source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"
+
+[[benchmarks]]
+name = "DeepSWE"
+score = 54.4
+metric = "score"
+source = "https://huggingface.co/deepseek-ai/DeepSeek-V4-Flash-0731"

providers/deepseek/models/deepseek-v4-flash.toml

@@ -1,14 +1,19 @@
+# DeepSeek-V4-Flash-0731 official API (public beta). Model id unchanged: deepseek-v4-flash.
+# https://api-docs.deepseek.com/updates (2026-07-31)
 # Reasoning tokens are billed at the output rate (no separate CoT price).
 # `completion_tokens_details.reasoning_tokens` is a subset of completion_tokens.
-# https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-07-31)
+# https://api-docs.deepseek.com/quick_start/pricing/ (accessed 2026-08-02)
+# Toggle: thinking.type = enabled|disabled
+# Effort: reasoning_effort = low|high|max (OpenAI); medium/xhigh map to high.
+# Anthropic: thinking.type / output_config.effort = low|high|max; budget ignored.
+# Responses: reasoning.effort = none|low|high|max (none disables thinking).
+# https://api-docs.deepseek.com/guides/thinking_mode
+# https://api-docs.deepseek.com/api/create-chat-completion
 name = "DeepSeek V4 Flash"
 description = "Fast DeepSeek V4 lane for economical reasoning, coding, and long-context work"
-# OpenAI: `thinking.type = enabled|disabled`, `reasoning_effort = high|max`.
-# Anthropic: `thinking.type`, `output_config.effort = high|max`; budget ignored.
-# https://api-docs.deepseek.com/api/create-chat-completion (accessed 2026-06-25)
 family = "deepseek-flash"
 release_date = "2026-04-24"
-last_updated = "2026-04-24"
+last_updated = "2026-07-31"
 attachment = false
 reasoning = true
 temperature = true
@@ -22,7 +27,7 @@ type = "toggle"
 
 [[reasoning_options]]
 type = "effort"
-values = ["high", "max"]
+values = ["low", "high", "max"]
 
 [interleaved]
 field = "reasoning_content"