Your signal. Your price.
Google’s Gemini 4 Carbon reportedly found 88% of vulnerabilities on SEC Bench Pro, exploited 55% of flaws on Exploit Bench, and wrote working patches for 71%. Alex said the model was not generally available and its benchmark position could reflect higher reasoning costs rather than a clear capability lead.
Alphabet shares fell 13% and lost $500 billion in market value despite revenue growth and a billion Gemini users; the panel pointed to weaker search growth and a $9 billion free-cash-flow burn. Dave argued that hyperscalers are borrowing heavily for AI while chip and memory shortages raise costs across consumer electronics.
Dhruv Gupta argues that broad professional search remains constrained: LinkedIn does not offer an MCP for extensive graph search, and his searches for New York startups returned weak results from Gemini and ChatGPT. Cosign wants contributed profiles and lists to make agent answers more useful and personalized.
Gemini 4 Carbon scored 59 on Artificial Analysis Index v5, a 9-point gain, and led six of eight highlighted benchmarks. It gained 8 points on Terminal Bench 4.0, 6 on Frontier Suite, 12 on Exploit Gym, and 7 on OSWorld.
Google released Gemini 4 Carbon two weeks after Argon but limited initial access to cyber defenders and trusted partners while it refined guardrails. Arena reviewer Peter Gostead said Carbon felt much better than Argon; Xiao Zhang described weeks of internal use for ambitious code changes and overnight experiments.
Frontier task costs vary sharply by model size and efficiency: GPT-6.1 Terra was cheapest, while Fable 5.1 was most expensive. Opus and Sonnet each cost about $10 per task; Astra cost $6.24, and Gemini 4 Carbon sat in the high-$7 range.
The hosts compare 36.5 trillion tokens processed by T3 Code users with Google’s 25 trillion Gemini Argon tokens in the same 72-hour window. T3 Code’s figure came only from users who enabled anonymous analytics, so the comparison is not like-for-like.
The hosts argue that Google’s benchmark strength has not translated into equally strong coding workflows, citing Gemini Argon’s perfect Skatebench score alongside slow product execution. They describe OpenAI Codex as incoherent and say better models can reduce the effort and fatigue of directing agents.
Genie 4 combines Google's models rather than operating alone: Gemini Omni anchors appearance and behavior, Gemini 3.8 Flash reasons over physical interactions, and Google audio models add sound. Users can steer the simulation live, alter weather or add objects, and generate more realistic physics, characters, speech, music, and effects.
Google says Gemini 3.8 Flash acts as a guiding agent inside Genie 4: it observes frames, breaks goals into steps, checks progress, and revises plans when the simulation changes. Google demonstrated the approach with Gemini 4 Argon guiding a robot through a chess task in a Genie simulation.
Prompt-generated games remain a longer-term prospect: Genie 4 is still a tech demo, and Gemini Ultra users can create simulations lasting up to two minutes. Investors reacted less sharply than after Genie 3's release, suggesting the market now sees a gap between impressive demos and disruption of the gaming business.
Genie 4 reinforced Google DeepMind's strength in multimodal systems despite weaker recent performance in language models. Nathaniel Whittemore said the release showed how combining different model architectures can produce a new kind of system, while Google's Gemini 4 Carbon was reportedly already in internal testing after Gemini 4 Argon.
Google's TPU business has gained customers beyond Anthropic: The Information reported new deals with OpenAI and Oracle, though the OpenAI terms were undisclosed and most existing TPU sales went to Anthropic. Google also reported processing about 25 trillion tokens in Gemini 4 Argon's first 72 hours, underscoring the reach of its distribution.