diff --git a/packages/opencode/src/tool/bash.ts b/packages/opencode/src/tool/bash.ts
index bf7c524941f..dd3d7c14121 100644
--- a/packages/opencode/src/tool/bash.ts
+++ b/packages/opencode/src/tool/bash.ts
@@ -58,7 +58,7 @@ export const BashTool = Tool.define("bash", async () => {
   return {
     description: DESCRIPTION.replaceAll("${directory}", Instance.directory)
       .replaceAll("${maxLines}", String(Truncate.MAX_LINES))
-      .replaceAll("${maxBytes}", String(Truncate.MAX_BYTES)),
+      .replaceAll("${maxTokens}", String(Truncate.MAX_TOKENS)),
     parameters: z.object({
       command: z.string().describe("The command to execute"),
       timeout: z.number().describe("Optional timeout in milliseconds").optional(),
diff --git a/packages/opencode/src/tool/bash.txt b/packages/opencode/src/tool/bash.txt
index 9fbc9fcf37e..71bab99afd1 100644
--- a/packages/opencode/src/tool/bash.txt
+++ b/packages/opencode/src/tool/bash.txt
@@ -24,7 +24,7 @@ Usage notes:
   - The command argument is required.
   - You can specify an optional timeout in milliseconds. If not specified, commands will time out after 120000ms (2 minutes).
   - It is very helpful if you write a clear, concise description of what this command does in 5-10 words.
-  - If the output exceeds ${maxLines} lines or ${maxBytes} bytes, it will be truncated and the full output will be written to a file. You can use Read with offset/limit to read specific sections or Grep to search the full content. Because of this, you do NOT need to use `head`, `tail`, or other truncation commands to limit output - just run the command directly.
+  - If the output exceeds ${maxLines} lines or ${maxTokens} tokens, it will be truncated and the full output will be written to a file. You can use Read with offset/limit to read specific sections or Grep to search the full content. Because of this, you do NOT need to use `head`, `tail`, or other truncation commands to limit output - just run the command directly.
 
   - Avoid using Bash with the `find`, `grep`, `cat`, `head`, `tail`, `sed`, `awk`, or `echo` commands, unless explicitly instructed or when these commands are truly necessary for the task. Instead, always prefer using the dedicated tools for these commands:
     - File search: Use Glob (NOT find or ls)
diff --git a/packages/opencode/src/tool/read.ts b/packages/opencode/src/tool/read.ts
index 3b1484cbc0f..8b26b7a3753 100644
--- a/packages/opencode/src/tool/read.ts
+++ b/packages/opencode/src/tool/read.ts
@@ -8,10 +8,11 @@ import DESCRIPTION from "./read.txt"
 import { Instance } from "../project/instance"
 import { Identifier } from "../id/id"
 import { assertExternalDirectory } from "./external-directory"
+import { Token } from "../util/token"
 
 const DEFAULT_READ_LIMIT = 2000
 const MAX_LINE_LENGTH = 2000
-const MAX_BYTES = 50 * 1024
+const MAX_TOKENS = 10_000
 
 export const ReadTool = Tool.define("read", {
   description: DESCRIPTION,
@@ -94,17 +95,17 @@ export const ReadTool = Tool.define("read", {
     const lines = await file.text().then((text) => text.split("\n"))
 
     const raw: string[] = []
-    let bytes = 0
-    let truncatedByBytes = false
+    let tokens = 0
+    let truncatedByTokens = false
     for (let i = offset; i < Math.min(lines.length, offset + limit); i++) {
       const line = lines[i].length > MAX_LINE_LENGTH ? lines[i].substring(0, MAX_LINE_LENGTH) + "..." : lines[i]
-      const size = Buffer.byteLength(line, "utf-8") + (raw.length > 0 ? 1 : 0)
-      if (bytes + size > MAX_BYTES) {
-        truncatedByBytes = true
+      const size = Token.estimate(line) + (raw.length > 0 ? 1 : 0)
+      if (tokens + size > MAX_TOKENS) {
+        truncatedByTokens = true
         break
       }
       raw.push(line)
-      bytes += size
+      tokens += size
     }
 
     const content = raw.map((line, index) => {
@@ -118,10 +119,10 @@ export const ReadTool = Tool.define("read", {
     const totalLines = lines.length
     const lastReadLine = offset + raw.length
     const hasMoreLines = totalLines > lastReadLine
-    const truncated = hasMoreLines || truncatedByBytes
+    const truncated = hasMoreLines || truncatedByTokens
 
-    if (truncatedByBytes) {
-      output += `\n\n(Output truncated at ${MAX_BYTES} bytes. Use 'offset' parameter to read beyond line ${lastReadLine})`
+    if (truncatedByTokens) {
+      output += `\n\n(Output truncated at ${MAX_TOKENS} tokens. Use 'offset' parameter to read beyond line ${lastReadLine})`
     } else if (hasMoreLines) {
       output += `\n\n(File has more lines. Use 'offset' parameter to read beyond line ${lastReadLine})`
     } else {
diff --git a/packages/opencode/src/tool/truncation.ts b/packages/opencode/src/tool/truncation.ts
index 84e799c1310..61dc4079a59 100644
--- a/packages/opencode/src/tool/truncation.ts
+++ b/packages/opencode/src/tool/truncation.ts
@@ -3,12 +3,13 @@ import path from "path"
 import { Global } from "../global"
 import { Identifier } from "../id/id"
 import { PermissionNext } from "../permission/next"
+import { Token } from "../util/token"
 import type { Agent } from "../agent/agent"
 import { Scheduler } from "../scheduler"
 
 export namespace Truncate {
   export const MAX_LINES = 2000
-  export const MAX_BYTES = 50 * 1024
+  export const MAX_TOKENS = 10_000
   export const DIR = path.join(Global.Path.data, "tool-output")
   export const GLOB = path.join(DIR, "*")
   const RETENTION_MS = 7 * 24 * 60 * 60 * 1000 // 7 days
@@ -18,7 +19,7 @@ export namespace Truncate {
 
   export interface Options {
     maxLines?: number
-    maxBytes?: number
+    maxTokens?: number
     direction?: "head" | "tail"
   }
 
@@ -49,44 +50,44 @@ export namespace Truncate {
 
   export async function output(text: string, options: Options = {}, agent?: Agent.Info): Promise<Result> {
     const maxLines = options.maxLines ?? MAX_LINES
-    const maxBytes = options.maxBytes ?? MAX_BYTES
+    const maxTokens = options.maxTokens ?? MAX_TOKENS
     const direction = options.direction ?? "head"
     const lines = text.split("\n")
-    const totalBytes = Buffer.byteLength(text, "utf-8")
+    const totalTokens = Token.estimate(text)
 
-    if (lines.length <= maxLines && totalBytes <= maxBytes) {
+    if (lines.length <= maxLines && totalTokens <= maxTokens) {
       return { content: text, truncated: false }
     }
 
     const out: string[] = []
     let i = 0
-    let bytes = 0
-    let hitBytes = false
+    let tokens = 0
+    let hitTokens = false
 
     if (direction === "head") {
       for (i = 0; i < lines.length && i < maxLines; i++) {
-        const size = Buffer.byteLength(lines[i], "utf-8") + (i > 0 ? 1 : 0)
-        if (bytes + size > maxBytes) {
-          hitBytes = true
+        const size = Token.estimate(lines[i]) + (out.length > 0 ? 1 : 0)
+        if (tokens + size > maxTokens) {
+          hitTokens = true
           break
         }
         out.push(lines[i])
-        bytes += size
+        tokens += size
       }
     } else {
       for (i = lines.length - 1; i >= 0 && out.length < maxLines; i--) {
-        const size = Buffer.byteLength(lines[i], "utf-8") + (out.length > 0 ? 1 : 0)
-        if (bytes + size > maxBytes) {
-          hitBytes = true
+        const size = Token.estimate(lines[i]) + (out.length > 0 ? 1 : 0)
+        if (tokens + size > maxTokens) {
+          hitTokens = true
           break
         }
         out.unshift(lines[i])
-        bytes += size
+        tokens += size
       }
     }
 
-    const removed = hitBytes ? totalBytes - bytes : lines.length - out.length
-    const unit = hitBytes ? "bytes" : "lines"
+    const removed = hitTokens ? totalTokens - tokens : lines.length - out.length
+    const unit = hitTokens ? "tokens" : "lines"
     const preview = out.join("\n")
 
     const id = Identifier.ascending("tool")
diff --git a/packages/opencode/test/tool/bash.test.ts b/packages/opencode/test/tool/bash.test.ts
index 750ff8193e9..e8749bc5540 100644
--- a/packages/opencode/test/tool/bash.test.ts
+++ b/packages/opencode/test/tool/bash.test.ts
@@ -253,16 +253,17 @@ describe("tool.bash truncation", () => {
     })
   })
 
-  test("truncates output exceeding byte limit", async () => {
+  test("truncates output exceeding token limit", async () => {
     await Instance.provide({
       directory: projectRoot,
       fn: async () => {
         const bash = await BashTool.init()
-        const byteCount = Truncate.MAX_BYTES + 10000
+        const tokenCount = Truncate.MAX_TOKENS + 1000
+        const byteCount = tokenCount * 4
         const result = await bash.execute(
           {
             command: `head -c ${byteCount} /dev/zero | tr '\\0' 'a'`,
-            description: "Generate bytes exceeding limit",
+            description: "Generate tokens exceeding limit",
           },
           ctx,
         )
diff --git a/packages/opencode/test/tool/truncation.test.ts b/packages/opencode/test/tool/truncation.test.ts
index 09222f279fa..3551a8bd930 100644
--- a/packages/opencode/test/tool/truncation.test.ts
+++ b/packages/opencode/test/tool/truncation.test.ts
@@ -1,6 +1,7 @@
 import { describe, test, expect, afterAll } from "bun:test"
 import { Truncate } from "../../src/tool/truncation"
 import { Identifier } from "../../src/id/id"
+import { Token } from "../../src/util/token"
 import fs from "fs/promises"
 import path from "path"
 
@@ -8,7 +9,7 @@ const FIXTURES_DIR = path.join(import.meta.dir, "fixtures")
 
 describe("Truncate", () => {
   describe("output", () => {
-    test("truncates large json file by bytes", async () => {
+    test("truncates large json file by tokens", async () => {
       const content = await Bun.file(path.join(FIXTURES_DIR, "models-api.json")).text()
       const result = await Truncate.output(content)
 
@@ -33,9 +34,9 @@ describe("Truncate", () => {
       expect(result.content).toContain("...90 lines truncated...")
     })
 
-    test("truncates by byte count", async () => {
+    test("truncates by token count", async () => {
       const content = "a".repeat(1000)
-      const result = await Truncate.output(content, { maxBytes: 100 })
+      const result = await Truncate.output(content, { maxTokens: 100 })
 
       expect(result.truncated).toBe(true)
       expect(result.content).toContain("truncated...")
@@ -63,18 +64,18 @@ describe("Truncate", () => {
       expect(result.content).not.toContain("line0")
     })
 
-    test("uses default MAX_LINES and MAX_BYTES", () => {
+    test("uses default MAX_LINES and MAX_TOKENS", () => {
       expect(Truncate.MAX_LINES).toBe(2000)
-      expect(Truncate.MAX_BYTES).toBe(50 * 1024)
+      expect(Truncate.MAX_TOKENS).toBe(10_000)
     })
 
-    test("large single-line file truncates with byte message", async () => {
+    test("large single-line file truncates with token message", async () => {
       const content = await Bun.file(path.join(FIXTURES_DIR, "models-api.json")).text()
       const result = await Truncate.output(content)
 
       expect(result.truncated).toBe(true)
-      expect(result.content).toContain("bytes truncated...")
-      expect(Buffer.byteLength(content, "utf-8")).toBeGreaterThan(Truncate.MAX_BYTES)
+      expect(result.content).toContain("tokens truncated...")
+      expect(Token.estimate(content)).toBeGreaterThan(Truncate.MAX_TOKENS)
     })
 
     test("writes full output to file when truncated", async () => {