As The Geek Learns AI Engineering Digestfrom As The Geek Learns

← Back to this issue

Monday, August 24, 2026

Frontier 4 stories

DeepSeek-V4-Flash-Vision-Exp

Experimental vision-understanding model on the DeepSeek API (deepseek-v4-flash-vision-exp). Official docs say it matches official DeepSeek-V4-Flash on pure-text and is a “significant leap” over V4-Flash on visual-understanding agent tasks; they describe multimodal agent capabilities as “close to Opus-4.8.” Images via base64, HTTPS URL, or Files API; billed as input tokens (up to 384 tokens each) at V4-Flash rates. Changelog benches (DeepSeek Harness, max effort) include Terminal Bench 2.1 83.9 and DeepSWE 59.3. vendor

NewsUpdatesVision guide

OpenAI: GPT-5.6 Sol API/credit price cut

Banner update on the GPT-5.6 launch post: “OpenAI dropped the API and credit pricing of GPT‑5.6 Sol by over 20% for the next 3 months.” The post does not state the new dollar rates. Original Sol list prices on the same page (July 9 launch copy): $5 / 1M input and $30 / 1M output (pre-cut).

GPT-5.6 post

OpenAI Ultrafast preview for GPT-5.6 Sol

New API service tier (not a new weights family): Ultrafast, powered by Cerebras, described as running GPT-5.6 Sol up to 14× faster than Standard, “up to 750 output tokens per second.” Limited preview for a select customer group; access to expand with capacity. vendor

Previewing Ultrafast

xAI Grok Voice Think Fast 2.0

Next-generation speech-to-speech voice model. API id grok-voice-think-fast-2.0. grok-voice-latest routes to 2.0 starting 2026-08-05. Price stated as $0.08 / min of audio. Table is “per Artificial Analysis”: Overall AA Speech-to-Speech Quality Index 82.9% vs 1.0’s 75.7%. xAI also claims transcription WER 1.5–2.0× better than named vendors; no raw WER in the prose. vendor

LaunchRelease notes