> ## Documentation Index
> Fetch the complete documentation index at: https://docs.tensormesh.ai/llms.txt
> Use this file to discover all available pages before exploring further.
# Changelog
> Latest updates, new features, and improvements to Tensormesh Inference.
## July 20, 2026
**External Storage Plan Limits** — Fixed an issue where External Storage plan limits were applied per model instead of to your bucket as a whole, letting total cached data exceed your plan's capacity. Your bucket is now shared evenly across serverless models, so each model is capped at its equal share. ([External Storage](/external-storage))
***
## July 6, 2026
**New Serverless Model: GLM-5.2-NVFP4**
`lukealonso/GLM-5.2-NVFP4` is now available for serverless inference, replacing `lukealonso/GLM-5.1-NVFP4-MTP`. This updated GLM-5.2 release brings improved performance over its predecessor while retaining the same 128K context window and NVFP4 quantization for efficient, high-throughput inference. ([Serverless Inference](/serverless-inference))
***
**Observability Metrics**
Three new charts are now live on **Operations → Observability**:
* **Time to First Token** — average latency from request submission to the first token returned
* **Input Throughput** — prompt tokens processed per second across your active requests
* **Output Throughput** — generated tokens per second, the primary driver of streaming response speed
These metrics appear per model alongside the existing cache charts. ([Observability](/observability))
***
## June 22, 2026
**Observability Dashboard**
**Operations → Observability** is a new per-model LMCache metrics dashboard. Filter by model and time range, and track:
* **Cache Hit Rate** — fraction of tokens served from cache at \$0.00
* **Cache Token Flows** — CPU hit tokens, CPU read / write / evict, and L2 (External Storage) read / write / evict throughput in tokens
Export any view as CSV for further analysis. ([Observability](/observability))
**KV Cache Calculator**
A new KV Cache Calculator is available on **Operations → Storage**. Select a model, set your context length and data type, and see exactly how many gigabytes of KV cache that conversation will consume — broken down step-by-step. Use it to estimate your bucket needs before subscribing to an External Storage plan. ([External Storage](/external-storage#kv-cache-calculator))
**Tensormesh Agent**
An AI assistant is now built into the platform. Click the assistant button in the bottom-right corner of any page to open a chat panel powered by Tensormesh. It includes quick-start prompts and guides, follow-up suggestions after each response, and an expand mode for longer conversations.
**Billing Address Required**
A billing address is now required alongside a payment method before you can add credit on the platform. When adding a card, you'll be prompted to complete your billing address if one isn't on file. You can update it anytime from **Management → Billing → Payment Methods**. ([Billing & Usage](/billing-and-savings#managing-payment-methods))
***
## May 27, 2026
**New Serverless Models**
Ten models are now available for serverless inference:
| Model | Family | Parameters | Context |
| ---------------------------- | -------------- | ---------- | ------- |
| DeepSeek-V4-Flash | DeepSeek-V4 | 284B | 1M |
| gemma-4-31B-it | Google Gemma 4 | 30.7B | 256K |
| GLM-5.1-NVFP4-MTP | GLM-5.1 | 433B | 128K |
| MiniMax-M2.5 | MiniMaxAI | 228B | 196K |
| Kimi-K2.6 | Kimi | 1T | 256K |
| gpt-oss-120b | OpenAI gpt-oss | 116B | 131K |
| gpt-oss-20b | OpenAI gpt-oss | 20B | 131K |
| Qwen3.5-397B-A17B-FP8 | Qwen3.5 | 397B | 262K |
| Qwen3.6-27B-FP8 | Qwen3.6 | 27B | 262K |
| Qwen3-Coder-30B-A3B-Instruct | Qwen3 Coder | 30.5B | 262K |
See [Serverless Inference](/serverless-inference#available-models) for more details.
**External Storage**
External Storage gives your serverless models a persistent KV cache bucket — so context is remembered across requests and sessions. Subscribe to Bronze, Silver, or Gold plans from **Operations → Storage** to increase bucket capacity and drive a higher fraction of your tokens to \$0. ([External Storage](/external-storage))
**Reserved GPU Clusters**
On-demand GPU deployments have been removed from the platform. If you need dedicated GPU capacity for high-volume production workloads, contact us through **Deploy → Reserved** to request a reserved cluster. The Tensormesh team will work with you on a tailored capacity plan. ([Reserved Deployments](/reserved-deployments))
**Overview Page** — The Overview page has been updated with quick actions linking to Serverless, Demos, Claude Code, and Codex CLI. The stats shows Cache Hit Rate, Models Called, Monthly Spending, and Account Balance. A cached tokens chart visualizes your cache performance over time, and a serverless model catalog is shown at the bottom for quick access.
**Cache Savings** — **Management → Cache Savings** now shows estimated savings, a stacked area chart of input spend vs. cache savings over time, and a top-models-by-savings table.
**SDK & CLI** — The SDK and CLI have been updated to focus on serverless inference.
**New Docs Pages** — Dedicated documentation for [External Storage](/external-storage), [Serverless Usage](/serverless-usage), [Cache Savings](/cache-savings), and a [Glossary](/glossary) of key terms.
***
## April 29, 2026
**Account Deletion** — You can now delete your own account from **Management → Account**. Account deletion requires email confirmation and is permanent.
**Billing Transaction Details** — Transaction details on the Billing page now show a deeper breakdown.
***
## April 15, 2026
**Serverless Inference**
Run models via API with no infrastructure to manage. Pay-per-token pricing with \$0 for cached tokens. Track per-model token usage and costs under **Operations → Serverless Usage**. The API is OpenAI-compatible — point any existing SDK to `https://serverless.tensormesh.ai`. ([Serverless Inference](/serverless-inference))
**Tensormesh Demos**
A new **Demos** section with interactive benchmarks. Navigate to **Operations → Demos** to run live inference demos and observe KV cache acceleration across TTFT, E2E latency, and inter-token latency.
**CLI Documentation**
A new **CLI** tab in the docs with guides and a full command reference for the `tm` CLI tool — covering installation, authentication, inference, model management, billing, and admin workflows. ([CLI](/cli/index))
**Python SDK**
A new **Python SDK** tab with guides for the `tensormesh` package — covering sync and async clients, inference, control plane resources, and migration from OpenAI/Fireworks. ([Python SDK](/sdk/index))
**API & SDK Reference Documentation**
Full interactive API & SDK docs for all Tensormesh endpoints with an in-browser playground and code examples in cURL, Python, and JavaScript. ([API & SDK Reference](/api-reference/introduction))
**Email Notification Preferences** — Toggle email notifications for deployment updates from **Management → Account**.
**Support Ticket Attachments** — You can now attach files when creating support tickets.
**Quick Actions on Dashboard** — Quick action cards on the Overview page for faster navigation to common operations.
***
## March 17, 2026
**MiniMax-M2.5 on Serverless**
`MiniMaxAI/MiniMax-M2.5` is now available for serverless inference. Built on a 228B-parameter Mixture-of-Experts architecture with a 196K context window, it excels at advanced reasoning, coding, and building autonomous systems that combine tool orchestration with large-scale information processing.
**Cost Saving Breakdown** — Cache Savings page now displays a step-by-step savings breakdown with calculation formulas.
**Browser Notifications** — Notification preferences can now be configured from **Management → Account**.