Google Gemini 4 Neon Costs $3 per Million Input Tokens

Google’s Gemini 4 Neon emerges as a lower-cost reasoning model for coding, research and production agents, positioned between Gemini 4 Carbon and Hydrogen. Google lists standard pricing at $3 per million input tokens and $15 per million output tokens, with cached input at $0.15, and a 1 million-token context window. Neon is available through the paid Gemini API, Google Cloud Console and, for eligible subscribers, Google Antigravity; its weights remain closed. Google also describes a Vertex AI route through its Gen AI SDK and offers Provisioned Throughput for reserved capacity.

Benchmark results put Neon near the top of several comparisons, though results vary by test. It scored 61 on the Artificial Analysis Intelligence Index and 59 on its Coding Agent Index, and achieved 95.2% on GraphWalks. Other reports place it second on Vals at 70.87% and second on Arena’s Agent Arena Index at 1,336; one reports first place on the Coding Agent Index. Neon also scored 76% on AutoBaxBench, 51.26% on WebExploitBench v1.2 and 90.5% on SEC-bench Pro. Google says it scored 70.7% on CWE-bench v2, 3.4 points above Carbon, and matched frontier performance on ExploitGym using half Carbon’s tokens. Neon for cyber is slated for trusted defenders through the Fairwind Program, not general access yet.

Anthropic figures in the AI market story from two different points in time: one article says the company has decided to pursue an IPO and is valued above $20 billion, while another reports a first-day stock-price doubling that put its valuation above $107 billion. The reports describe investor scrutiny of its growth plans, transparency, AI safety and data-center concerns, alongside expectations of strong interest. CEO Dario di Pardo says Anthropic is committed to addressing criticism and improving safety; the stock has been volatile in recent days.

Amazon’s AWS is helping customers move AI beyond demonstrations through Forward Deployed Engineering engagements built around 45-day sprints. Teams establish technical foundations, workflows and governance, while customers keep systems in their own AWS accounts. At Marsh, an initial use case reached production about two months after work began, and some targeted tasks became up to 70% more efficient. AWS says it can train customer engineers to manage and expand the systems after an engagement.

AI deployment also appears in consumer, healthcare and workplace reports. Walmart says AI agents that recommend products, personalize shopping and improve efficiency helped support stronger e-commerce performance; it reported third-quarter adjusted EPS of $0.65 and raised its full-year forecast to $2.84–$2.89 from $2.34–$2.40. In Fresno, Saint Agnes, Community Regional and Kaiser Permanente hospitals use AI to draft visit notes, with clinicians reviewing final records and patients able to opt out. Logitech sees AI-enabled products as a longer-term opportunity as it grows business-to-business sales, while P&G’s fiscal 2027 plan targets 1%–3% organic sales growth and uses AI-enabled consumer insights alongside productivity and supply-chain work.

Nvidia’s RTX Spark brings local AI computing to a desktop, combining a 20-core Grace CPU and Blackwell RTX GPU with 6,144 CUDA cores for up to 1 petaflop of AI performance. Nvidia says it can run models with up to 120 billion parameters, render 3D scenes larger than 90GB and edit 4:2:2 video in real time. Adobe Premiere and Photoshop and Blender 5.3 support related features, though software availability remains a constraint: Substance 3D Painter 12.2 is still in public beta on Arm and more native apps are pending.

Other tests suggest AI results depend on the surrounding tools as well as the model. Across 5,779 agent runs on 878 websites, a plain model fetched an average of 1.3 pages and answered 2.49 of eight questions, versus Cursor’s 8.2 pages and 4.25 answers; Gemini CLI answered 5.98 questions, but was tested on only 59 sites. The runs took place from October 1 to November 17, 2026. Separately, Samaya’s AI forecasts beat Wall Street analyst estimates in 40 of 50 third-quarter company comparisons, using Bloomberg consensus figures and public information. Morgan Stanley co-president Dan Simkowitz called AI the biggest investment-capital spending cycle of his lifetime and highlighted capital markets’ role in financing infrastructure.

Key Takeaways

  • Google Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, offers a 1 million-token context window and is available through paid developer and cloud channels.
  • Neon scored strongly across coding, reasoning and cybersecurity tests; Google says its cyber version will initially go to trusted defenders through the Fairwind Program.
  • Reports describe Anthropic’s prospective IPO at an earlier valuation above $20 billion and a later first-day trading valuation above $107 billion, with safety, transparency and data-center concerns drawing scrutiny.
  • AWS Forward Deployed Engineering uses 45-day sprints to help customers operationalize AI; Marsh put an initial use case into production in about two months, with some tasks up to 70% more efficient.
  • Walmart raised full-year adjusted EPS guidance to $2.84–$2.89 after reporting $0.65 for the third quarter, citing e-commerce capabilities and AI agents.
  • Three Fresno hospital systems let patients opt out of AI-assisted documentation; clinicians review and approve the resulting records.
  • Nvidia says RTX Spark delivers up to 1 petaflop of AI performance and can run models with up to 120 billion parameters; software support is still developing, including on Arm.
  • In a 878-site agent test, Cursor answered more questions than a plain model, while Gemini CLI had the highest reported answer count but a smaller test sample.
  • Samaya’s forecasts were more accurate than Wall Street consensus in 40 of 50 third-quarter earnings comparisons.
  • Logitech identifies AI-enabled products as a longer-term opportunity; P&G targets 1%–3% organic sales growth in fiscal 2027, and Morgan Stanley says AI is driving an unusually large infrastructure investment cycle.

Google Gemini 4 Neon promises fast reasoning at lower cost

Google’s Vertex AI model page describes Gemini 4 Neon as a fast reasoning model for coding, reasoning and production agents. Google says Neon builds on Gemini 4 Hydrogen and delivers quality close to Gemini 4 Carbon at Flash tier speed and cost. The model is available on the Gemini platform through Google’s Gen AI SDK, including Google AI Studio and the Agent Platform, with support for Python, Go, JavaScript, Java and .NET. Its model ID is gemini-4-neon, and Provisioned Throughput is available for businesses that want to reserve capacity.

Gemini 4 Neon targets cybersecurity work at lower cost

Google DeepMind says Gemini 4 Neon costs $3 per million input tokens and $15 per million output tokens, with cached input at $0.15, and has a 1 million-token context window. In company-reported tests, Neon for cyber scored 76% on AutoBaxBench, 51.26% on WebExploitBench v1.2, and 90.5% on SEC-bench Pro. It also scored 70.7% on CWE-bench v2 for patching, 3.4% above Carbon, and matched frontier performance on ExploitGym while using half Carbon’s tokens. Neon with real-time safeguards at its highest thinking level had a 0.0% attack success rate on Gray Swan’s IPI benchmark. Google plans to offer Neon for cyber to trusted defenders through the Fairwind Program in the coming days; it is not yet open to everyone.

Google pricing listings point to a Gemini 4 Neon launch

Pricing entries for Google’s Gemini 4 Neon have appeared, though the model remains unannounced. The listings show standard prices of $3 per million input tokens and $15 per million output tokens, with batch, flex and contributor rates at $1.50 and $7.50. Cached input is listed at $0.15, while priority processing costs $5.40 for input and $27 for output per million tokens. Around the same time, Google DeepMind product lead Logan Kilpatrick posted “Gemini” on X, but neither his post nor Google linked it to Neon. The listings do not guarantee a launch date.

Gemini 4 Neon scores above Carbon on an AI intelligence index

Gemini 4 Neon scored 61 on the Artificial Analysis Intelligence Index at the xhigh setting, one point above Gemini 4 Carbon at high, and ranked third overall. Neon costs $6.44 to run that index, about 17% less than Carbon at its standard price, though Neon uses about 15% more output tokens per task. On the Artificial Analysis Coding Agent Index, Neon scored 59, ahead of Carbon’s 55 and several competing systems. Artificial Analysis also reported gains in visual reasoning and presentation quality, including a 170 Elo increase on the AA-Briefcase Presentation Quality measure. Neon has a 1 million-token context window and accepts text, images, video and speech, with text output.

Google DeepMind releases Gemini 4 Neon for coding and research

Google DeepMind released Gemini 4 Neon as a reasoning model between Gemini 4 Carbon and Gemini 4 Hydrogen, aimed at coding agents, knowledge work and science tasks. It has a 1 million-token context window, a default output limit of 64,000 tokens, and supports up to 1 million tokens through background long decode. Google lists prices of $3 per million input tokens, $0.15 for cached input, and $15 per million output tokens. Google-reported results include a score of 61 on the Artificial Analysis Intelligence Index, first place on its Coding Agent Index, and 95.2% on the GraphWalks long-context test. Developers can use Neon through the paid Gemini API tier, Google Antigravity for paid Google AI Pro and Ultra subscribers, and Google Cloud Console; its weights are closed.

Gemini 4 Neon ranks near the top across major AI benchmarks

Google’s Gemini 4 Neon placed second on the Vals Index with 70.87% accuracy, just behind Anthropic’s Claude Fable 5.2 and ahead of Claude Opus 5.5. It also ranked second on Arena’s Agent Arena Index, scoring 1,336, one point above Opus 5.5. Neon cost $7.40 per Vals test and finished in about 31 minutes, less than half the cost and time reported for Sonnet 5.5. It led the models shown on CursorBench 4.1 at comparable costs and ranked first on Artificial Analysis’s Coding Agent Index, but placed behind Fable 5.2 and Sonnet 5.5 on AA-Briefcase.

Logitech growth draws support from business expansion and AI products

A Morgan Stanley summary says Logitech International’s growth is supported by market share gains, premium products, gaming and business to business expansion. Regional weakness, supply issues and cost pressures are near term challenges. AI enabled products are among the company’s longer term opportunities. Logitech also plans to shift toward an even mix of consumer and business to business sales.

P&G targets modest sales growth as it invests in productivity and AI

Procter & Gamble’s fiscal 2027 guidance calls for organic sales growth of 1% to 3%. Its long term plan targets mid to high single digit growth in core earnings per share and adjusted free cash flow productivity of more than 90%. The company says productivity, AI enabled consumer insights, supply chain upgrades and innovation will support its growth plans.

Anthropic reportedly prepares for an IPO amid investor concerns

The article says Anthropic has decided to go public and has been valued at more than $20 billion. Some experts question its transparency, growth strategy and valuation, while others expect the IPO to draw strong investor interest. Anthropic’s CEO said the company is on track to deliver significant growth and returns to investors. The article describes the IPO as a test of how investors view the company’s prospects and risks.

Anthropic’s record IPO puts AI safety and data center concerns in focus

Anthropic’s stock more than doubled on its first day of trading, bringing the company’s valuation to over $107 billion. Investors are weighing concerns about its AI safety approach and public anger toward data centers. Critics say its technology may not prevent errors and that the company is not doing enough to address risks. CEO Dario di Pardo said Anthropic is committed to addressing criticism and making its AI safer, while the article reports that the stock has been volatile in recent days.

AWS helps businesses move AI from demos into production

AWS Forward Deployed Engineering teams work with customers to put AI tools into business use through 45-day sprints. The teams build the technical foundation, workflows, and governance, while customers keep control of the systems and run them in their own AWS accounts. At Marsh, the first AI use case reached production about two months after work began, and some targeted tasks became up to 70 percent more efficient. AWS says customers can train their engineers to manage and expand the systems after the engagement ends.

Fresno hospitals let patients choose whether AI helps document visits

Saint Agnes Medical Center, Community Regional Medical Center, and Kaiser Permanente Fresno Medical Center all use AI to help document patient visits, but patients can choose whether to use it. The tools can draft clinical notes and other written material, while clinicians review and approve the final records. Saint Agnes says its AI can also help draft portal responses and plain-language test summaries. The hospitals say they protect patient information under privacy rules, and their descriptions explain that audio or recordings are stored only temporarily or are not saved, depending on the tool.

Nvidia RTX Spark brings high power AI computing to a local desktop

NVIDIA says its RTX Spark combines a 20-core Grace CPU with a Blackwell RTX GPU and 6,144 CUDA cores for up to 1 petaflop of AI performance. It can run models with up to 120 billion parameters locally, render 3D scenes larger than 90GB, and edit 4:2:2 video in real time. The system includes 10-gigabit Ethernet, while Adobe Premiere and Photoshop and Blender 5.3 add support for related features. Software coverage remains a limit, with Substance 3D Painter 12.2 still in public beta on Arm and more native apps pending.

AI agent tools found more website answers but took longer

A test of 5,779 AI agent runs across 878 websites found that the software around a model affects how much information it can find. The plain model fetched an average of 1.3 pages and answered 2.49 of eight questions, while Cursor answered 4.25 and fetched 8.2 pages on average. Gemini CLI answered 5.98 questions, but it was tested on only 59 sites. The runs took place from October 1 to November 17, 2026, and the article advises businesses to keep key facts reachable within a few links of their homepages.

Samaya AI forecasts beat Wall Street estimates in 40 of 50 cases

Fortune and financial AI startup Samaya compared AI forecasts with Wall Street analyst estimates for 50 companies reporting third-quarter earnings from October 26 to November 5. Samaya’s predictions were more accurate in 40 cases, using estimates for each company’s key measure, such as earnings per share or revenue. The test covered ten sectors and compared the AI forecasts with analyst consensus figures from Bloomberg. Samaya CEO Maithra Raghu said the startup relied on public information, while noting that analyst research could be added to its training data in the future.

Walmart lifts earnings outlook as AI agents improve shopping

Walmart reported third-quarter adjusted earnings per share of $0.65, above estimates. It raised its full-year EPS guidance to $2.84 to $2.89 from $2.34 to $2.40. The company said stronger e-commerce capabilities and AI agents helped drive the increase. The agents recommend products, personalize shopping, reduce costs, and improve efficiency. Walmart’s stock price and market value rose after the improved guidance.

Morgan Stanley executive calls AI the biggest investment cycle of our lifetimes

Morgan Stanley co-president Dan Simkowitz said AI is driving the biggest investment capital spending cycle of our lifetimes. He spoke with CNBC’s Lisa Kim at the Morgan Stanley Asia Pacific Summit. Simkowitz highlighted the growing role of capital markets in financing AI infrastructure.

Sources

NOTE:

This news brief was generated using AI technology (including, but not limited to, Google Gemini API, Llama, Grok, and Mistral) from aggregated news articles, with minimal to no human editing/review. It is provided for informational purposes only and may contain inaccuracies or biases. This is not financial, investment, or professional advice. If you have any questions or concerns, please verify all information with the linked original articles in the Sources section below.

Google Gemini 4 Neon AI reasoning models AI coding agents Gemini API Vertex AI AI benchmarks cybersecurity AI Anthropic IPO AI safety AWS Forward Deployed Engineering enterprise AI deployment AI agents AI in retail AI healthcare documentation Nvidia RTX Spark local AI computing AI software support AI agent evaluation AI financial forecasting AI infrastructure investment

Comments

Loading...