TokForge 1.3.8

TokForge measures what your own phone does well and uses it, and new users get a much simpler start.

Last verified October 13, 2026

Beta from October 10, 2026, on the Google Play beta track and as the Discord APK; replaced by 1.3.8.1 on October 11, 2026. Version 1.3.7 was never released, so this page lists everything that changed since 1.3.6.1. Step-by-step help: the how-to and the terminal guide.

• Getting started
• Your phone picks the route
• GPU support, phone by phone
• Memory
• Faster on the CPU
• Speed features, measured per phone
• Chat and prompts
• Agents tab
• Terminal (opt-in, first version)
• Settings and UI
• Backup and restore
• Other fixes
• Known limits in 1.3.8

Getting started

• Easier first run. Tell TokForge what you want to do (chat, writing, code or pictures) and it picks one model that fits your phone, with its size, a speed estimate and a Wi-Fi note.
• What it picks. Chat and Code get a Qwen3.5 model sized to your phone's memory (0.8B, 2B or 4B). Writing gets Qwen3 4B Abliterated on phones with 8 GB or more.
• Simpler model list. Models are sorted into Best for your phone, Fits and Too big, and the technical details sit behind one Advanced switch. If you already use TokForge, your model list stays as it was.
• One Advanced switch for the whole app hides engine words (MNN, GGUF, quant names, OpenCL, Vulkan, NPU). If you are upgrading, it starts switched on, so nothing moves for you.
• Plain descriptions. Catalog descriptions are rewritten in plain words in all 8 languages, and none of them promises the GPU any more, because your phone decides that.
• New Qwen3 1.7B build in the catalog. A copy of the old one you already installed stays.

Your phone picks the route

• GPU check while idle. For the usual automatic GPU setups, a new model starts right away and TokForge compares the GPU's answers with the CPU's once, while the app sits idle. A GPU setup you pick yourself gets that check before it starts.
• A fair check. It judges only the words the CPU is sure about, so a near tie does not fail a GPU that answers correctly. If the check fails, the model stays on the CPU and its card says why in plain words.
• First check in its own process. The first GPU check of an MNN model runs in a separate helper, so a GPU problem during it does not close the app.
• CPU or GPU, by measurement. TokForge times the CPU, the GPU and a mix of both (GPU reads the prompt and CPU writes the reply, or the other way round) for the model you loaded, and keeps whichever measured faster on your phone. Settings (Advanced mode) > Speed & engine > Performance > Measured on your phone shows what it found. Each route and speed feature there has its result in plain words and Off, Auto and On, and one switch covers all of them.
• Rechecks when something changes. Results are tied to the app version, the system update, the GPU driver and the model file, so a change to any of them gets a fresh check.
• Long prompts on the GPU. The recommended Qwen3.5 model can read long prompts on the GPU while the CPU writes the reply, where your phone's own check passes and the measurement wins. Everywhere else it stays on the CPU.
• Steadier speed checks. The automatic check times whole replies, a few times each, before it picks, and a phone that runs hot gets a shorter check. Settings (Advanced mode) > Speed & engine > Performance > Pause background checks holds the automatic checks while it is on.
• Checks stay out of your way. Automatic checks run only while the app sits idle in the foreground, the phone is cool, the battery is not low (or the phone is charging) and power saving is off. They wait while a download runs and stop as soon as you type, switch models or leave the app.
• Clearer GPU status. The route line in AutoForge > Tune says whether its GPU route was checked, is still waiting for its check, or ran without the output check.
• AutoForge keeps your measured route. AutoForge no longer switches off the routes your phone measured. It still tunes threads, precision, cache and context size, and its result names the route in use.

GPU support, phone by phone

• Every GPU route is used only where your phone's own check passes and the measurement wins. Otherwise the model stays on the CPU.
• Snapdragon 8 Gen 3 and Snapdragon 8 Elite Gen 5. MNN can measure the "GPU prompt, CPU reply" mix, and MNN on OpenCL no longer gives wrong or empty replies when a chat continues.
• Newer Snapdragon GPUs. GGUF on Vulkan follows rules for each GPU family that keep its answers correct, and is offered only where those rules apply, with crash protection while it loads.
• Pixel 9. An MNN OpenCL crash is fixed, and MNN can measure the "GPU prompt, CPU reply" mix with tuned Vulkan settings.
• Any other GPU. GGUF on Vulkan reads your phone's real GPU name, so GPUs that were never on our list get measured too, with crash protection. A Vulkan pick you make yourself is honoured there after its check.
• Pixel 10 and Snapdragon 865. These phones stay on the CPU for now, even when you pick Vulkan yourself.
• More models on the mix. Llama 3.2 3B on MNN can now measure the "GPU prompt, CPU reply" mix too.
• Plain reasons. If you pick a mix your phone cannot run (for example, not enough memory or the phone's GPU memory limit), TokForge tells you why and runs the model on the CPU.
• Recovery. If the GPU stalls, TokForge moves that model to the CPU and, once nothing of yours is running, restarts itself in the background to free the GPU.

Memory

• One copy of the model for CPU and GPU. For GGUF models, the "CPU prompt, GPU reply" route can share one copy of the model between the CPU and the GPU instead of loading two, where your phone's own check passes and the measurement wins. The route line says when both share one copy.
• Weights counted once. While a GGUF model loads, TokForge lets go of the file pages it has already copied, so loading takes less memory.
• Speed checks respect memory limits. Each route the automatic check would try must first fit your phone's free memory and the system's per-app limits, or it is skipped.
• Per-app memory limits. On phones whose system caps how much memory one app may use, GPU mixes, speed checks and catalog badges respect that cap.

Faster on the CPU

• Long prompts. On phones whose processor supports it, such as the Pixel 9 and the Xiaomi 14T Pro, the first reply to a long prompt comes sooner.
• Replies. GGUF models write their replies faster on the CPU on many phones, after a fix to how the processor's cores wait for each other.
• Pixel 9. GGUF replies use four CPU threads instead of two.

Speed features, measured per phone

• Extra speed only where your phone shows a gain. Prompt lookup, multi-token prediction and the Llama draft pack each get a quick test on your phone, and TokForge turns them on only where a complete test shows a gain. Each has Off, Auto and On in Settings (Advanced mode) > Speed & engine > Performance > Measured on your phone.
• Speed boost on Snapdragon 8 Gen 3 and Snapdragon 8 Elite Gen 5. An optional download lets the phone's AI chip read your prompt, so first replies to long prompts come sooner with Qwen3 0.6B, 1.7B and 4B Abliterated. It stays off unless you turn on Faster long prompts (NPU) or tick it on the first-run card, checks itself on its first load, and stays on the CPU if that check fails. Its download waits for Wi-Fi.
• Other opt-ins. The Gemma 4 mix (GPU reads the prompt, CPU writes the reply) and Vulkan for picture generation stay off unless you turn them on in Settings (Advanced mode) > Speed & engine > Performance > Measured on your phone.

Chat and prompts

• Quicker follow-ups in Qwen chats (GGUF models, thinking off): TokForge reuses the earlier part of the chat instead of reading all of it again on every turn.
• First message in a new chat is kept exactly once, and it waits a moment for the chat's saved settings.
• Your whole message. MNN models keep your whole current message when it fits. A message too long to fit keeps its start and end, and the chat tells you.
• Long MNN chats keep track of positions correctly past about 2,000 tokens.
• No more "No model loaded" dead end. If a chat's model is not loaded but another local model is, the chat uses the loaded one and says so.
• Your message comes first. Background memory work waits while you are in a chat.

Agents tab

• See what each agent does. Each agent has its own run screen: type a request, watch its steps, stop it, and use the result. There are three starter presets (Look it up, Write or rewrite, Summarize), and your recent runs can be run again.
• Example requests. Each preset has examples you can tap. Look it up searches the web and lists its sources, Write or rewrite drafts text, and Summarize takes a file or pasted text.
• Use the result. Select the text, open the sources, Copy, Share or Save as file.
• Plain errors in your language, and leaving a running agent asks before it stops the run.

Terminal (opt-in, first version)

• Off by default. Settings Mode: Advanced, then Connections > Developer tools (Advanced) > Terminal (sandboxed). It turns on only after your phone passes its own check.
• What it is. A command shell that works in its own scratch folder inside TokForge. Its tools (toybox and curl) come with the app: there is no package manager, and it does not download programs to run.
• Internet is off unless you turn on Allow internet in the terminal.
• Commands in chat. Ask the model to run something, or type a line starting with $, and an approval shows the exact command first. With a remote chat model, the approval also says the output will be sent to it.
• Stop everything ends all commands at once. Ctrl+C stops the current command.
• The full guide is on the terminal page.

Settings and UI

• My Models lists the loaded model first, and the Settings screen adapts to wider screens.
• Effective backend sits right under the MNN and GGUF backend choices and names a CPU and GPU mix when one is in use.
• Model names say Abliterated, and the speed on the first-run card is an estimate.

Backup and restore

• Restore works on Android 15 and newer. Restore accepts good backups and still refuses a damaged one before your current data is touched.
• Backups include what you just wrote. Chats, messages and characters saved in the last minutes before a backup are included, and the counts match the file. Make a fresh backup after updating.

Other fixes

• The most common "TokForge isn't responding" stop is fixed.
• Scrolling and typing stay smooth while a GGUF model writes a reply on the processor.
• If the GPU runs out of memory during an MNN reply, the reply stops with an error instead of closing the app.
• Mistral 7B v0.3 and Gemma 3 1B (MNN): a broken tokenizer file is repaired when the model loads.
• The catalog's Fits badge asks the same memory question as loading, so a model too big for your phone says Too big.
• App notices no longer hide behind three-button navigation.

Known limits in 1.3.8

• After updating, each phone measures its routes once while idle and cool, which can take a while per model in the background. A phone that keeps running hot keeps its current setup.
• The terminal is a first version, available only where your phone passes its check. Android 11 phones cannot run it, and here-documents (cat <<EOF) fail: write files with echo or printf and run scripts with sh.
• Pixel 10 and Snapdragon 865 phones stay on the CPU for now. Large mixture-of-experts models may stay on the CPU too.
• Long MNN chats can drop their oldest messages sooner than they need to.
• Importing a chat export adds every chat again as a copy instead of merging.
• If the network drops during a model download, the download stops; tap Retry and it continues.

All versions and changelogs · Report a problem