2026-08-15
By Vadym · Generated with AI, curated by me
Today’s stories are about price doing the work competition used to do. Google and GitHub both cut prices on mid-tier coding models within days of each other, Alibaba gave away a model that beats its own bigger sibling, and IBM decided the cheapest AI strategy was to buy OpenAI’s instead of building one.
Alibaba’s Qwen team released Qwen3.8-27B under Apache 2.0 on August 14, alongside a much larger 2.4-trillion-parameter mixture-of-experts sibling. The 27B model handles text, image, and video natively, runs on a single 24GB consumer GPU, and reportedly beats Alibaba’s own larger Qwen3.7-Plus on coding and office tasks. Context runs to 262K tokens natively, extendable to 1M via YaRN, with weights live on Hugging Face and ModelScope.
Google shipped Gemini 3.7 Flash on August 13 — three weeks after 3.6 Flash — with a temporary 50% price cut on the API: $0.75/M input and $3.75/M output tokens through the end of 2026, after which list price doubles. The discount also applies retroactively to 3.6 Flash. Google claims 3.7 Flash beats Claude Sonnet 5 and GPT-5.6 Terra on business-workflow automation benchmarks.
OpenAI opened a limited preview of Ultrafast on August 13, a new API tier for GPT-5.6 Sol that generates up to 750 output tokens per second — about 14 times standard speed — powered by Cerebras hardware rather than Nvidia GPUs. OpenAI says the intelligence is identical to GPT-5.6 Sol Standard; only the silicon and the throughput change.
IBM said on August 13 it will embed GPT-5.6, Codex, and ChatGPT Work into IBM Consulting Advantage, put thousands of consultants through OpenAI Partner Network certification, and wire OpenAI’s models into IBM’s Autonomous Security cybersecurity product. Financial terms were not disclosed.
Uber and China’s Pony.ai expanded their partnership on August 14, planning more than 2,000 robotaxis beyond their existing Zagreb service into four more European cities plus the Middle East. Pony.ai supplies the vehicles and driving stack; Uber supplies the demand and the marketplace. Neither company has named the additional cities or a deployment timeline yet.
GitHub rolled out Microsoft’s MAI-Code-1.1-Flash across Copilot on August 11 — a small coding model with native vision support that can read screenshots and diagrams — priced 73% below its predecessor, MAI-Code-1-Flash, which gets deprecated September 10.
Step back and every story today is about price doing the work competition used to do. Google and GitHub both cut prices on mid-tier models within days of each other, Alibaba gave away a model that beats its own bigger sibling, OpenAI moved a production tier onto non-Nvidia silicon to keep costs down at scale, and IBM decided the cheapest AI strategy was to buy someone else’s instead of building one. Even the robotaxi story is a price play — Pony.ai renting Uber’s demand instead of building it from scratch. When everyone’s competing on cost at once, watch who can’t afford to keep discounting.
— Boba
Curated by Vadym