Public Company · Frontier Lab
ModelPublic CompanyMountain View
75/100
↗ +23.7 (7d)
Oct 28
Reports Q3 2026 earnings: a record $51.4B quarterly capex (more than double a year earlier) and Google Cloud revenue up 103% YoY, but discloses it will lean on rented third-party compute capacity in Q4 because its own buildout can't keep pace with demand, and warns 2027 capex will “increase significantly” again.
Oct 28
Unreleased Gemini 4 Carbon tops Artificial Analysis's revamped Intelligence Index v5 (score 59, first of a harder, mostly-private benchmark), beating every publicly available model, though early access testers report it running two to four times slower than predecessor Gemini 4 Argon and with a higher hallucination rate.
Oct 27
Partners with virtual-power-plant operator Voltus on a 1.17-gigawatt expansion across the PJM grid territory, days after FERC rejected PJM's plan to require data centers to accept mandatory curtailment before gaining new grid connections — a sign Google is betting on buying flexible capacity rather than waiting on a curtailment mandate regulators just struck down.
Oct 21
Google Public Sector and Accenture Federal Services launch "Gemini Launchpads," a federal-agency agent program nearly identical to Accenture's May 2026 OpenAI federal partnership — Accenture is now running parallel, non-exclusive AI-agent programs with both labs inside the same government buyer.
Oct 18
Independent benchmark operators split sharply on Gemini 4 Argon: Vals AI ranks it first on a finance/legal/tax-weighted index, while Artificial Analysis ties it for eighth on a broader intelligence composite — as Bloomberg reports some Google employees privately call the model "benchmaxxed" on real-world coding despite its strong standardized scores.
Oct 17
Reportedly negotiating a voluntary "Standards Authority for Frontier AI" (SAFA) with OpenAI and Anthropic for third-party pre-deployment testing, even as new research shows the evaluation and monitoring infrastructure such testing would rely on can be tampered with or evaded by the models being assessed.
Oct 8
Releases EmbeddingGemma 2, a 740-million-parameter model that maps text, code, images, video, and audio into a single vector space and runs on a Pixel phone in under 600 MB, extending the original EmbeddingGemma's 20-million-download run beyond text-only retrieval. Released under Apache 2.0.
Oct 7
Agrees to buy nearly 3.6 gigawatts of power from Constellation Energy in a deal reported at $4.3B, funding nuclear plant uprates across three states as hyperscalers race to secure electricity ahead of chips.
Oct 6
DeepMind open-sources SynthID Bio, which embeds a detectable watermark in AI-designed protein sequences and structures without degrading their function, landing weeks after a Microsoft-led study showed AI design tools can slip dangerous DNA variants past the screening systems synthesis companies rely on.
Oct 5
Bilibili, a Chinese video platform, open-sources a family of translation models that beat Gemini 3.5 Flash Lite on standard benchmarks using a fraction of the active parameters — a small release with a bigger implication for narrow, task-built models.
Oct 1
Gemini 4 Argon, Google DeepMind's first new flagship model in nine months, beats GPT-6 Astra and Claude Opus 5.5 on most disclosed benchmarks, but the company routes it first to vetted cybersecurity defenders rather than the open market.
Sep 29
Agrees to pay roughly $900M to fund uprates squeezing about 96 additional megawatts out of four aging Vogtle and Hatch nuclear units in Georgia — a fraction of its data center appetite, but a template for buying capacity that already exists rather than waiting years for new reactors.