Latest Artificial Intelligence News & Technical Dispatches

Benchmark speculative decoding with SPEED-Bench data showing 2.9x wins at batch 1 collapsing at batch 32, plus vLLM tuning that holds gains in production.

Cover Crusoe $3B raise at $30B plus the $13B Jane Street GPU cloud deal, with gigawatt campuses, IPO signals, and a migration playbook for builders.

Cover AIUC $55M raise to scale AIUC-1 audits and agent insurance with 5000 adversarial sims, Big Four backing, and payouts reaching tens of millions.

Study agent rot with the geometric decay law showing collapse within 16 steps, plus decomposition and gate patterns that hold reliability in production.

Deploy prompt caching for agents with system-first layout, three tested strategies, and TTFT guardrails that cut API bills 80 percent in production.

TypeSafe AI exits stealth with $40M for Jev, a decision model returning typed judgments with calibrated confidence at sub-100ms for software agents.

K2 Horizon ships six Apache 2.0 models from 0.9B to 375B with training data, checkpoints and logs plus rare self-reported benchmark corrections verified.

Design KV cache for 1M-token agents with paged attention, prefix routing, MLA compression and FP8 quantization to cut memory 400x for production fleets.

Arcee AI raises Series B past $1B after building Trinity 400B MoE for $20M, funding next-gen models, DOE science work and proven open-model products.

Compare Fable 5.1 at 11% enterprise spend vs Opus 5 and Gemini 3.8 Flash with benchmark deltas and a proven routing gateway that cuts token costs 68%.

Kimi K2.8 Preview ships Sep 11 under the same kimi-for-coding ID with near-K3 speed, full 1M context for all tiers and adjustable thinking levels.

DeepSeek routes V4-Pro traffic to V4.1 Flash since Sep 14 with CED architecture, 90.9 GPQA scores and 30x lower cost. Full migration steps inside.

Give agents ColPali visual retrieval over document screenshots with late interaction matching and binary compression for tables that never survive parsing.

Sweep reasoning effort per task class with proven low high max tiers, default down from max thinking, and escalate on repair rates to cut spend 40%.

Atria Dawn Preview ships quiet MIT 744B MoE weights with 5 vendor-reported top scores. Independent evals, hosting dates and context checks still pending.