AI Launch Profile
DeepSeek V4-Flash-0731 launches in API beta with native Codex support
DeepSeek released DeepSeek-V4-Flash-0731 as the official V4-Flash API public beta. The checkpoint keeps the preview model’s architecture and size but adds new post-training, native Responses API support, and an official Codex configuration path. DeepSeek also published the 0731 weights in a public, ungated MIT-licensed Hugging Face repository.
At a glance
Launch Snapshot
- Company
- DeepSeek
- Launch date
- July 31, 2026
- Launch type
- Major Update, Model Release
- Category
- AI Agents, AI Coding Tools, AI Developer Tools, AI Models
- Audience
- AI Engineers, Developers, Engineering Teams
- Pricing
- $0.0028 per 1M cached input tokens; $0.14 per 1M uncached input tokens; $0.28 per 1M output tokens. DeepSeek says future peak-hour prices will be 2× regular rates, with no effective date announced.
- Free plan
- No
- API
- Yes
- Open weights/source
- Yes
Launch Context
Use these links to move from this record into the broader Launch Intelligence database.
Verification & Sources
- Status
- Verified
- Source links
- 9
- Freshness
- Verified July 31, 2026
- Last verified
- July 31, 2026
- Last updated
- July 31, 2026
Key source checks
Suggest a correction
Creator Coverage Next Steps
This launch has signals that may support demos, reviews, creator education, founder storytelling, or practical product explainers.
Launching an AI product that needs clear demos, creator education, and buyer trust? Sponsor a Kingy AI video or launch feature.
Kingy AI Take
Worth testing for cost-sensitive coding-agent workloads because it combines native Codex support, a one-million-token context window, very low API pricing and MIT-licensed weights. Treat the benchmark gains as vendor claims until DeepSeek releases its test harness and independent teams reproduce them.
Who it is for
Developers, AI engineers and engineering teams evaluating lower-cost coding agents through Codex CLI, the ChatGPT desktop app or the Codex extension for VS Code.
What feels promising
Native Responses API support removes a compatibility layer for Codex clients, while public MIT-licensed weights add a self-hosting path. DeepSeek reports broad gains across public agent benchmarks without changing the model architecture, while keeping uncached API input pricing at $0.14 per million tokens.
What feels unproven
Kingy.ai has not rerun the benchmarks, tested the model in a production repository or validated a self-hosted deployment. DeepSeek used an unreleased harness, maximum reasoning effort and specific sampling settings; two cited DSBench results come from internal test sets.
Traction notes
At review time on July 31, 2026, DeepSeek’s X announcement showed roughly 2.9 million views, while Cline’s benchmark comparison showed more than 500,000 views. These figures indicate attention, not independent validation.
Source-backed record