Google

Key Terms

• Token-based billing - charges based on the number of input and output tokens processed. One token is roughly 4 characters. Most Gemini 3.x models use a 200K-token threshold (requests at or below 200K input tokens billed at the standard rate, larger requests at a higher rate); Gemini 3.6 Flash and 3.5 Flash-Lite are billed at a flat rate across the 200K threshold. Source: Google – Pricing
• Prompt caching - stores frequently used prompt prefixes to reduce cost. Cached input tokens cost roughly 90% less than standard input tokens. Available for all Gemini 3.x and 2.5 models. Source: Google – Pricing
• Context window - the maximum tokens a model can process in one conversation. Gemini 3.6 Flash, 3.5 Flash, 3.1 Pro, and 3 Flash support a 1M token context window. Source: Google – Code Assist
• Antigravity CLI - Google's Go-based terminal AI agent that replaced Gemini CLI for free and individual users on June 18, 2026. It orchestrates multiple agents asynchronously and shares the same agent harness as the Antigravity desktop app. Source: Googleblog – An Important Update Transitioning Gemini Cli To Antigravity Cli
• Antigravity 2.0 - Google's agent-first development platform, announced at I/O 2026, including a standalone desktop app for orchestrating cohorts of autonomous AI agents. Source: Google – Sundar Pichai Io 2026
• Interactions API - Google's GA unified endpoint for Gemini models and agents, with server-side state, background execution, Managed Agents, and a typed "Steps" schema replacing the old role structure. It is the default for Google AI Studio, the Gemini API, and all documentation. Source: Google – Interactions Api General Availability
• Managed Agents - an Interactions API feature (GA) that provisions a remote Linux sandbox where an agent can reason, run code, browse, and manage files; the Antigravity agent ships as the default and custom agents are defined as markdown files. In July these gained environment hooks, free-tier access, budget controls, scheduled triggers, and an Environments API. Source: Google – Expanding Managed Agents Gemini Api 3 6 Flash Hooks
• Environment hooks - scripts (command or HTTP) that run before or after every tool call an agent makes inside its remote sandbox, enabling block/lint/audit of tool calls such as code_execution and write_file. New in July. Source: Google – Expanding Managed Agents Gemini Api 3 6 Flash Hooks
• Agent mode - a multi-step collaborative reasoning agent inside Gemini Code Assist chat that can edit multiple files, use full project context, call built-in tools, and integrate MCP servers. Available on both Standard and Enterprise. Source: Google – Code Assist
• Global vs Non-global pricing - some Flash models offer lower pricing when requests are served from Google's global compute pool, and 10% higher pricing when routed to a specific non-global region (e.g. 3.5 Flash-Lite non-global input is $0.33/MTok vs $0.30 global). Gemini 3.6 Flash is published at a single global rate only. Source: Google – Pricing
• Priority tier - a premium serving option with higher availability and lower latency at 1.8x the standard price, available for all Gemini 3.x models. Source: Google – Pricing
• Flex/Batch tier - a discounted serving option for non-urgent workloads at roughly 50% of standard pricing. Source: Google – Pricing

Latest Changes

Changes since the 2026-06 report. June watch items are verified first.

• Failed / still-pending (June watch-item): Gemini 3.5 Pro did NOT launch in July; it slipped again. Google's July 21 model launch post states "Gemini 3.5 Pro is currently testing with partners and we plan to make it broadly available as soon as it's ready," and confirms the team "have started our most ambitious pre-training run yet, for Gemini 4." A July 17 Los Angeles Times investigation (by Bloomberg's Julia Love and Davey Alba) reports Google is "months behind schedule on delivering Gemini 3.5 Pro" and that a June training-data update intended to improve coding "results were disappointing." No launch date and no API price have been announced. Sources: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber , Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers
• Confirmed (June watch-item): Consumer Gemini Code Assist on GitHub fully shut down July 17, 2026. The deprecation document still reads "Starting July 17, 2026, the consumer version is shut down, and all code review activities performed by the app end." The enterprise version is unaffected. The doc was last updated 2026-05-27 UTC. Source: Google – Consumer Code Review
• Partial (June watch-item): Antigravity CLI feature parity improved but no 1:1 parity matrix published. The Managed Agents layer gained environment hooks, free-tier access, budget controls (max_total_tokens), scheduled triggers (cron), and an Environments API for listing/inspecting/deleting sandbox sessions, but Google has still not published a parity matrix versus the legacy Gemini CLI. Source: Google – Expanding Managed Agents Gemini Api 3 6 Flash Hooks
• Confirmed (new July item): Gemini 3.6 Flash launched (July 21). Positioned as the new workhorse: better coding, knowledge work, and multimodal performance than 3.5 Flash, at lower output cost. Priced at $1.50 input / $7.50 output per MTok (Global, flat across the 200K threshold). Google cites the Artificial Analysis Index for a 17% reduction in output tokens versus 3.5 Flash, and up to 65% on DeepSWE (Datacurve). Available in the Gemini API, Google AI Studio, Android Studio, Google Antigravity, Gemini Enterprise Agent Platform, Gemini Enterprise app, and the Gemini app. Source: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber
• New model (GA): Gemini 3.5 Flash-Lite (July 21). The fastest, cheapest 3.5-class model at $0.30 input / $2.50 output per MTok (Global), running at 350 output tokens/sec per the Artificial Analysis Index, with built-in computer use as a client-side tool. It replaces the prior 3.1 Flash-Lite ($0.25/$1.50) as the efficient-tier option and outperforms 3 Flash on several agentic benchmarks. Source: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber
• New model (limited access): Gemini 3.5 Flash Cyber in CodeMender. A cybersecurity-specialized model built on 3.5 Flash, fine-tuned for finding and fixing vulnerabilities, paired with the CodeMender code-security agent. Available exclusively to governments and trusted partners via CodeMender as a limited-access pilot, not via the public API. Source: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber
• Product update: Managed Agents expanded (July 28). The antigravity-preview-05-2026 agent now defaults to Gemini 3.6 Flash (no code change required) with explicit model selection (3.6 Flash, 3.5 Flash, 3.5 Flash-Lite). Added environment hooks (block/lint/audit tool calls in-sandbox), free-tier access for projects without active billing, budget controls via max_total_tokens that pause execution at the cap and return status: "incomplete" with state preserved, scheduled cron triggers, and an Environments API with a 7-day TTL. Source: Google – Expanding Managed Agents Gemini Api 3 6 Flash Hooks
• Internal reporting: LA Times/Bloomberg investigation (July 17). Reports Google is "months behind" on 3.5 Pro; that Google Cloud, Google DeepMind, and the Android team are all building competing AI coding tools; that Chief AI Architect Koray Kavukcuoglu is working to unite internal tools (with a DeepMind coding team led by Sebastian Borgeaud); that internal Anthropic Claude access is now restricted to cutting-edge research and high-priority projects; that engineers hit compute-capacity constraints when using AI; and that Google states 75% of code at the company is now AI-generated and reviewed to production. Alphabet shares slipped as much as 3.2% on the report. Source: Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers
• Security signal: sandbox-escape research (July 20). BleepingComputer reported that Cursor, Codex, Gemini CLI, and Antigravity (among others) were affected by sandbox-escape findings, a cross-tool agent-security reminder relevant to anyone running these CLIs. Source: Bleepingcomputer – Cursor Codex Gemini Cli Antigravity Hit By Sandbox Escapes
• Product update: Gemini Spark Chrome integration (July 30). Gemini Spark gained Chrome auto-browse (using logged-in accounts and saved passwords, with prompt-injection defenses and human handoff for payments) and rolled out to Google AI Pro subscribers in over 160 additional countries. This is a consumer/agent-platform item, not a Code Assist change, but it reflects Google's cross-product agent direction. Source: Google – Gemini Spark Updates July 2026
• Pricing verified: Code Assist Standard stays $22.80/user/month ($19 annual) and Enterprise $54/user/month ($45 annual). The Vertex AI API pricing page now lists Gemini 3.6 Flash and Gemini 3.5 Flash-Lite alongside the existing 3.5 Flash, 3.1 Pro Preview, 3 Flash Preview, and 2.5 models. Sources: Google – Pricing , Google – Pricing

Plans

Hourly list prices (how Google Cloud bills commitments): Standard $0.031232877/hour (monthly) and $0.026027397/hour (12-month); Enterprise $0.073972603/hour (monthly) and $0.061643836/hour (12-month). At 730 hours/month these resolve to the $22.80/$19 Standard and $54/$45 Enterprise figures above. Source: Google – Pricing

Terms explained:

• Code customization - tailors code suggestions to an organization's private codebases. Enterprise-only. Source: Google – Pricing
• Gemini Cloud Assist - AI-powered infrastructure design, diagnostics, cost optimization, and troubleshooting. Preview, at no additional charge for Enterprise customers. Source: Google – Pricing
• Gen AI indemnification - Google covers legal costs if Gemini Code Assist output infringes third-party copyright. Available on both Standard and Enterprise. Source: Google – Protecting Customers With Generative Ai Indemnification

Source: Google – Pricing , Google – Code Assist

API Pricing

Gemini 3.6 Flash and Gemini 3.5 Flash-Lite pricing were re-verified on the live Vertex AI pricing page on July 31, 2026. 3.6 Flash is published at a single Global rate (no non-global tier, flat across the 200K threshold). 3.5 Flash-Lite carries Global and Non-global variants. 3.5 Flash, 3.1 Pro Preview, and the older models are unchanged from June.

Standard Pricing

Note: the prior 3.1 Flash-Lite ($0.25/$1.50) is superseded by the GA Gemini 3.5 Flash-Lite on the current lineup.

Flex/Batch Pricing (~50% discount)

Priority Pricing (1.8x standard)

Notes:

• Gemini 3.5 Pro is not yet priced because the model is not GA. Google says it is "currently testing with partners" with no date or price. Source: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber
• Gemini 3.5 Flash Cyber is available only to governments and trusted partners via CodeMender (limited-access pilot), so no public per-token API price is published. Source: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber
• Vertex AI Model Garden also sells partner models (e.g. Claude Sonnet 5), with standard pricing beginning September 1, 2026. Source: Google – Pricing

Source: Google – Pricing

Model Performance / Benchmarks

Google published fresh benchmark numbers for Gemini 3.6 Flash and Gemini 3.5 Flash-Lite on July 21, both reported as gains over their immediate predecessors. Gemini 3.5 Pro still has no published scores because it is not GA.

Google's comparison charts are again presented as images rather than structured tabular data, so exact head-to-head scores versus Claude Opus 4.8, Sonnet 5, GPT-5.6, and GLM-5.2 on the same benchmark are not extractable from official Google sources. The Artificial Analysis Index (a third-party aggregator) is the source Google itself cites for token-efficiency and speed claims. Source: Artificialanalysis – Gemini 3 6 Flash

Latest News

Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber Launch (July 21, 2026)

Google launched Gemini 3.6 Flash as the new efficient workhorse at $1.50/$7.50 per MTok (17% fewer output tokens than 3.5 Flash per the Artificial Analysis Index, with higher DeepSWE, MLE Bench, OSWorld, and GDPval scores), promoted Gemini 3.5 Flash-Lite to GA at $0.30/$2.50 per MTok (350 tokens/sec, replacing 3.1 Flash-Lite), and introduced Gemini 3.5 Flash Cyber in CodeMender as a limited-access cybersecurity pilot for governments and trusted partners. The post reiterates that Gemini 3.5 Pro "is currently testing with partners" and that pre-training for Gemini 4 has begun. Sources: Google – Gemini 3 6 Flash 3 5 Flash Lite 3 5 Flash Cyber , Artificialanalysis – Gemini 3 6 Flash

Managed Agents Get Hooks, Free Tier, Budget Controls (July 28, 2026)

The Managed Agents surface in the Interactions API now defaults to Gemini 3.6 Flash and added environment hooks (pre/post tool-call scripts that can block, lint, or audit sandbox actions), free-tier access for projects without billing, max_total_tokens budget caps that pause execution at the limit, scheduled cron triggers, and an Environments API. AI-native investment bank OffDeal is cited using post-tool hooks for in-sandbox image verification. Source: Google – Expanding Managed Agents Gemini Api 3 6 Flash Hooks

LA Times/Bloomberg: 3.5 Pro "Months Behind," Internal Friction (July 17, 2026)

An investigation by Julia Love and Davey Alba (Bloomberg, published in the LA Times) reports Gemini 3.5 Pro is "months behind schedule," that a June training-data refresh for coding produced "disappointing" results, that Google Cloud, DeepMind, and Android are all building competing coding tools, that internal Claude access is now restricted to cutting-edge research and high-priority teams, and that engineers hit compute-capacity constraints. Google responded that it is "currently testing 3.5 Pro, an upgraded Flash model, and other models with partners" and is "productively engaged with the U.S. government on model testing." Shares slipped as much as 3.2%. Source: Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers

Consumer Gemini Code Assist on GitHub Reaches Hard Shutdown (July 17, 2026)

The consumer version of Gemini Code Assist on GitHub fully shut down on July 17, 2026, ending all code review activities, as the deprecation document specified. The enterprise version is unaffected. Source: Google – Consumer Code Review

Gemini Spark Integrates with Chrome (July 30, 2026)

Gemini Spark added Chrome auto-browse (using logged-in accounts and saved passwords with prompt-injection defenses and human handoff for payments) and expanded to Google AI Pro subscribers in over 160 additional countries. This is a consumer/agent-platform item that signals Google's cross-product agent direction rather than a Code Assist change. Source: Google – Gemini Spark Updates July 2026

Agent Security: Sandbox Escapes Hit Gemini CLI and Antigravity (July 20, 2026)

BleepingComputer reported that Cursor, Codex, Gemini CLI, and Antigravity were among the coding agents affected by sandbox-escape findings, a cross-tool security reminder for teams running these CLIs against untrusted code. Source: Bleepingcomputer – Cursor Codex Gemini Cli Antigravity Hit By Sandbox Escapes

Community Signals

The 3.5 Pro delay and internal friction drew negative attention on Hacker News

The Los Angeles Times (Bloomberg wire) investigation into Google's Gemini delay, submitted to Hacker News (24 points, 6 comments), reinforced the community's concern that Google is falling behind Anthropic and OpenAI specifically in coding, the first major enterprise AI use case.

• HN submission (24 points, 6 comments): "Google's Gemini delay: Coding stumbles, clashing teams and frustrated engineers." News – Item
• Primary source article (permalink for all quotes below): Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers

Customer sentiment on 3.5 Flash is split, and one customer has already left

The LA Times/Bloomberg piece captured two on-record customer views of the still-current Gemini 3.5 Flash, and they diverge sharply. A Figma product manager found a sweet spot, while the Platzi CEO found the model occupying an awkward middle ground and has shifted spend to Anthropic.

• Rodrigo Davies, product manager at Figma (on adding 3.5 Flash to the Figma agent): the model "hit a sweet spot of speed and quality." Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers
• Freddy Vega, CEO and founder of Platzi: "It is more expensive than Google's previous 3.1 Flash model, yet slower, and it remains far less capable than premium offerings from competitors. It often struggles with structured data." Vega added that for tasks needing a balance of speed and reasoning, "his team has shifted away from Google to one of Anthropic's mid-tier models." Latimes – Inside Googles Gemini Delay Coding Stumbles Clashing Teams Frustrated Engineers

Agent sandbox security touched Gemini CLI and Antigravity

The BleepingComputer sandbox-escape story was submitted to Hacker News multiple times (top submission 8 points) and lists Gemini CLI and Antigravity alongside Cursor and Codex as affected agents, a signal that Google's CLI agents share the same class of agent-security exposure as competitors.

• HN submission (8 points): "Cursor, Codex, Gemini CLI, Antigravity hit by sandbox escapes." News – Item
• Primary source (permalink): Bleepingcomputer – Cursor Codex Gemini Cli Antigravity Hit By Sandbox Escapes

Enterprise Readiness

Terms explained:

• VPC Service Controls - lets enterprises define a security perimeter around Google Cloud resources to prevent data exfiltration. Source: Google – Code Assist
• SOC certifications - Gemini Code Assist has achieved SOC 1/2/3, ISO/IEC 27001, 27017, 27018, and 27701 certifications. Source: Google – Code Assist

Transparency Gaps