Program changelog

Google’s Cloud Billing release notes list spend cap budgets as generally available for a limited set of eligible services. Unlike alerts-only budgets, they pause new requests when estimated spend exceeds the budget amount, which gives trial users a real stop switch. Source

Google’s AI Studio billing docs now say that for Cloud Billing accounts opened after March 2, 2026 the $300 Welcome credit can’t be used for Gemini API or AI Studio usage. It still applies to other Google Cloud products, including Google models on Vertex AI. Source

Google published a new version of the Free Trial supplemental terms, last modified February 12, 2026. The terms keep the $300 credit, the three-month expiry, the eight concurrent Compute Engine core limit and the ban on cryptocurrency mining. Nothing got better, nothing got worse. Source

Google announced on its Keyword blog that Google Developer Program premium benefits are folded into Google AI Pro and Ultra subscriptions, including $10 a month (Pro) or $100 a month (Ultra) in Google Cloud credits usable on Vertex AI, Cloud Run and the Gemini API. Source

Google’s Vertex AI generative AI release notes record that Agent Engine Runtime reached general availability with a new free tier, after charging had begun on November 6, 2025 in five regions. Sessions, Memory Bank and Code Execution began billing on January 28, 2026. Source

Google’s API changelog records that Gemini 2.5 Pro and 2.5 Flash are deprecated. From December 1, 2026 their requests route to Gemini 3.8 Flash and Gemini 3.5 Flash-Lite and are billed at those models’ rates. Access to 2.5 models had been limited to past users since September 18, 2026. Source

The API changelog adds Flex and Priority inference options beside Standard on April 1, 2026. Google’s price list now shows separate per-token prices per tier, with Flex input on Gemini 3.8 Flash at $0.375 per million against $0.75 Standard and $1.35 Priority. Source

The API changelog lists Prepay and Postpay billing plans for Gemini API usage in AI Studio. Google’s billing docs set a $5 minimum and $5,000 maximum prepay balance and say all keys on a billing account stop with HTTP 402 at $0. New Gemini API terms took effect the same day. Source

The API changelog records revamped usage tiers and new spend caps at the billing account level, following project-level spend caps on March 12, 2026. Google’s rate limits docs now list Tier 1 at $250, Tier 2 at $2,000 and Tier 3 at $20,000 to $100,000 or more per month. Source

Google’s AI Studio billing docs state that Cloud Billing accounts opened after March 2, 2026 can’t use the $300 Welcome credit for Gemini API or AI Studio usage. Credits granted before that date could be spent on Gemini until their 90-day expiry. Source

OpenAI’s API changelog records that eligible ChatGPT Pro and Business subscribers can claim up to $500 per month in free API credits. Pro grants match the tier price; Business grants pool $20 per Standard and $100 per Premium seat. Source

OpenAI’s API changelog adds a 12-hour prompt cache retention option for GPT-5.6 and later models in the Responses API. Cache writes at that lifetime cost 2.5x the input token price, versus 1.25x for the default 30-minute cache. Source

The API changelog entry simplifies usage tiers to Build, Launch and Grow. OpenAI’s rate limits guide now sets thresholds at $5, $100 and $500 in total credit purchases with monthly usage limits of $500, $5,000 and $200,000. Source

OpenAI’s API changelog notes GPT-5.6 Luna now costs 80% less and Terra 20% less. Fast mode replaces Priority processing as the low-latency option, with Sol’s Fast rate at twice the standard price. Source

The API changelog records that the dashboard gained hard monthly spend limits that stop requests at a chosen dollar amount for an organization or a single project. These sit alongside, not in place of, the approved usage limit for each tier. Source

An update added to Anthropic’s October 6 expansion article on claude.com says the current cohort is full and new applications aren’t being accepted. The program page now points founders to the newsletter for the next cohort, and existing members keep their benefits. Source

Anthropic’s launch post prices Haiku 5.5 at $0.10 per million input tokens and $0.50 output for prompts up to 100,000 tokens, rising to $0.50 and $2.50 above that threshold. It ships on AWS, Google Cloud and Azure as well as the first-party API. Source

Anthropic’s announcement on claude.com adds a free year of Claude Team with up to five Premium seats for companies new to Team, a one-time $1,000 API credit on approval, and partner offers worth up to $45,000. Startups founded in the last five years or funded in the last two may apply. Source

The Sonnet 5.5 launch post holds the price at $2 per million input and $10 output, the same as Sonnet 5, while cache reads fall to $0.10, half of Sonnet 5. Anthropic says the model costs up to 30% less per task because it uses fewer tokens. Source

Anthropic’s launch post prices Opus 5.5 at $4 per million input tokens and $20 output, with cache reads at $0.20, 60% below Opus 5. Fast mode runs at $8 and $40. Anthropic says it costs 40% less than Opus 5 on typical workloads at default settings. Source

A note on Amazon’s Bedrock price list says that starting October 7, 2026 the cache read price for Claude Sonnet 5.5 is 50% lower than before. Base input and output rates didn’t change, so the saving lands on applications that reuse long prompts. Source

An AWS What’s New post dated March 25, 2026 adds Aurora PostgreSQL Serverless to the Free plan. It restates the $100 sign-up credit and the extra $100 earned through activities, and tells you to select the Free Plan during account sign-up. Source

AWS’s What’s New post of November 18, 2025 gives Bedrock two tiers beside Standard. Flex is priced at a discount and deprioritized at peak demand, while Priority costs more and delivers up to 25% higher output tokens per second on supported models. Source

The AWS News Blog announced that accounts created on or after July 15, 2025 receive $100 at sign-up plus $20 per activity up to $200, with a Free plan that ends at six months. Accounts created earlier keep the legacy 12-month free tier. Source

Microsoft’s startup FAQ on Learn, updated October 30, 2026, lists a $200 milestone for 90 days, $4,800 for 180 days after business verification, a $25,000 milestone with a two-year window, and up to $150,000 over time. Source

Microsoft’s Azure blog announced general availability of GPT-5.5 in Microsoft Foundry for the following day, priced at $5 input, $0.50 cached input and $30 output per million tokens, with GPT-5.5 Pro at $30 input and $180 output. Source

The Azure blog post for Ignite 2025 presented the platform as Microsoft Foundry with more than 11,000 models and announced Claude Sonnet 4.5, Opus 4.1 and Haiku 4.5 in the catalog. Microsoft Learn now lists Azure AI Foundry and Azure AI Services as the previous names. Source

The Azure blog of August 7, 2025 made GPT-5, GPT-5 mini, GPT-5 nano and GPT-5 chat generally available, with a model router that Microsoft said can cut inference costs by up to 60% by routing prompts across the family. Source

Microsoft Learn’s changes page says that from July 2, 2025 Microsoft for Startups split into an investor-referral track starting at $100,000 in credits and a self-serve offer of up to $5,000 for startups new to Azure, retiring the Founders Hub brand. Source

xAI’s news post for Grok 4.7 prices the model from $2 per million input tokens and $6 per million output tokens on the Grok API, with a fast variant at twice the price limited to Cursor and Grok Build. The price list adds $0.50 cached input and a 20% batch discount. Source

xAI’s announcement put Grok 4.6 in the API, OpenRouter, Vercel and Cloudflare at $2 per million input and $6 per million output tokens. The post offered doubled included usage for one week in Cursor and Grok Build, but no API credit. Source

The Grok 4.5 news post set pricing at $2 per million input tokens and $6 per million output tokens and listed availability in Grok Build, Cursor and the console. Free Grok 4.5 usage for a limited time applied to Grok Build and Cursor only, not to the API. Source

xAI’s rate limits page states that a team’s tier is based on cumulative API spend since January 1, 2026, with Tier 1 at $50, Tier 2 at $250, Tier 3 at $1,000 and Tier 4 at $5,000. Tiers never downgrade, and increases beyond Tier 4 go through the console or sales. Source

OpenRouter’s announcement introduced Private Endpoints, which let organizations route their own fine-tuned models, dedicated deployments and reserved capacity through OpenRouter with public providers as fallback. It’s an Enterprise add-on, usage is billed by the underlying provider, and each organization gets 10 endpoints by default. Source

OpenRouter’s announcement says batched requests generally cost 50% of a provider’s normal per-token price across more than 70 models, completed within a 24-hour window. The beta median was 7 minutes. Chat, responses, messages and embeddings are supported; audio, video and web search are not. Source

The announcement introduced us.openrouter.ai and eu.openrouter.ai endpoints that keep prompts and completions inside the chosen region, available on the Business and Enterprise plans. OpenRouter’s price list now shows Business as a pay-as-you-go plan with an 8% fee on credit purchases against 5.5% on Standard. Source

OpenRouter’s blog post said the company agreed to be acquired by Stripe, subject to closing conditions, and that nothing about existing integrations, the product, name or roadmap changes. The post made no changes to pricing, fees or credits, and the pass-through pricing model stayed as documented. Source

DigitalOcean’s inference release notes say spend alerts can be set on Serverless Inference model access keys, with an option to pause the key when spend reaches 100% of its budget. It’s the first hard cap on the platform. Source

The billing release notes say Serverless Inference and Managed Agents are now funded from a separate Inference and Agents balance that you can prepay and set to reload automatically, keeping model calls off the monthly invoice. Source

A DigitalOcean blog post announced new on-demand and 12-month reserved prices for select NVIDIA and AMD GPU Droplets effective August 1, 2026. Customers with locked-rate contracts keep their rate until renewal. Source

The inference release notes state that Serverless Inference now requires a positive prepaid account balance before requests are accepted, and that access is suspended when the balance reaches $0. Source

The Startups Terms of Service, last updated May 11, 2026, replace the former Hatch terms. Credits run 12 months from acceptance with a monthly consumption cap, and unused credits are forfeited at the end of the term. Source

Groq’s changelog says MiniMax M2.5 and Qwen3-VL 32B are available only to Enterprise customers, who must contact their Groq account team for access. Free and Developer accounts can’t call them. Source

Groq’s changelog records the policy-following gpt-oss-safeguard-20b model going live with automatic prompt caching that halves the price of cached input tokens, matching the discount on the other GPT-OSS models. Source

The changelog entry says automatic prompt caching is live for GPT-OSS 120B, halving cached input costs and excluding cached tokens from rate limits, which effectively raises the Free plan tokens per minute ceiling. Source

Groq’s changelog lists Kimi K2-0905 at $1.00 per million input tokens and $3.00 per million output tokens, with prompt caching saving up to 50% on cached portions of the prompt. Source

The changelog records the first prompt caching release, on Kimi K2, with a 50% saving on cached prompt portions and no additional fee, and promises more models to follow. GPT-OSS models were added over the next two months. Source