Selection window: only news published within the last 24 hours by absolute (UTC) time — cutoff 2026-07-30 21:07 to 2026-07-31 21:07 UTC. Stories from the daytime of 30 July UTC that yesterday’s brief already covered (OpenAI’s GPT-5.6 price cut, Google’s Gemini Robotics 2, Anthropic’s test-breach disclosure, and the Microsoft/Apple/Amazon earnings) fall before the cutoff and are excluded today. This window is essentially 31 July UTC news.
DeepSeek ships “V4-Flash-0731”: Opus-4.8-class performance from 284B parameters, at rock-bottom prices
On 31 July, DeepSeek announced DeepSeek-V4-Flash-0731, a re-trained build of its smaller V4-Flash model, in its API changelog as a public beta. It keeps the same architecture and size as April’s preview — 284B total parameters (13B activated per token), a 1M-token context window and 384K max output — and only re-ran post-training to sharpen agentic, coding and multi-step tool-use abilities. It beat DeepSeek’s own higher-tier V4-Pro-Preview on all nine published agent and coding benchmarks; on the agentic-coding metric Terminal-Bench 2.1 it jumped from the preview’s 61.8 to 82.7, above Pro-Preview’s 72.1. Observers noted that this 284B model approaches the performance of Anthropic’s Opus 4.8, believed to be a multi-trillion-parameter system. Pricing is $0.14 per million input tokens and $0.28 per million output, and it newly supports OpenAI’s Responses API format and Codex compatibility.
Reference: Digital Watch · TechTimes (07/31) · officechai
The three-way AI price war escalates: DeepSeek’s $0.14/$0.28 undercuts OpenAI’s fresh cut by ~30%
V4-Flash-0731’s pricing neutralized, within hours, the cut OpenAI had made a day earlier (30 July), when it dropped GPT-5.6 Luna by 80% to $0.20 input / $1.20 output. DeepSeek’s new rate slices roughly 30% under OpenAI’s just-discounted Luna tier and runs several times cheaper than Anthropic’s Haiku 4.5. With Moonshot AI also scaling up (roughly 20,000 new NVIDIA GPUs) to reinforce the open-weight camp, a three-way low-cost fight (OpenAI vs. DeepSeek vs. Moonshot/open-weights) for developer spend is now in full swing on the high-volume, routine-workload tier. The key implication: separate from the flagship performance race, token prices are collapsing across providers simultaneously precisely in the “volume tier” where most enterprise spend actually lands.
Reference: wccftech · Startup Fortune
Today’s summary
Over the last 24 hours (UTC), AI news centered on DeepSeek’s new small-model build, V4-Flash-0731. At 284B parameters it claims Opus-4.8-class performance at a striking $0.14 input / $0.28 output, immediately answering OpenAI’s 80% Luna cut from the day before. Combined with Moonshot AI’s GPU build-out, a three-way price war over “volume-tier” token pricing has become the defining thread of this cycle.
Daily AI Brief 
Leave a comment