[Regression] Codex repeatedly reprocesses massive cached context in long sessions, causing severe latency, timeouts, JSONL growth, and excessive credit usage

What version of Codex CLI is running?

codex-cli 0.145.0

What subscription do you have?

ChatGPT Plus

Which model were you using?

gpt-5.4 with xhigh reasoning

What platform is your computer?

Microsoft Windows 11 Home | Version 10.0.26200 | 64 bit

What terminal emulator and version are you using (if applicable)?

Codex CLI TUI launched from PowerShell on Windows PowerShell version: 5.1.26100.8875

Codex doctor report

{
  "schemaVersion": 1,
  "generatedAt": "1784823909s since unix epoch",
  "overallStatus": "warning",
  "codexVersion": "0.145.0",
  "checks": {
    "app_server.status": {
      "id": "app_server.status",
      "category": "app-server",
      "status": "ok",
      "summary": "background server is not running",
      "details": {
        "control socket": "C:\\Users\\<USER>\\.codex\\app-server-control\\app-server-control.sock",
        "daemon state dir": "C:\\Users\\<USER>\\.codex\\app-server-daemon",
        "mode": "ephemeral",
        "pid file": "C:\\Users\\<USER>\\.codex\\app-server-daemon\\app-server.pid (missing)",
        "settings": "C:\\Users\\<USER>\\.codex\\app-server-daemon\\settings.json (missing)",
        "status": "not running",
        "update-loop pid file": "C:\\Users\\<USER>\\.codex\\app-server-daemon\\app-server-updater.pid (missing)"
      },
      "remediation": null,
      "durationMs": 0
    },
    "auth.credentials": {
      "id": "auth.credentials",
      "category": "auth",
      "status": "ok",
      "summary": "auth is configured",
      "details": {
        "auth env vars present": "<redacted>",
        "auth file": "C:\\Users\\<USER>\\.codex\\auth.json",
        "auth storage mode": "<redacted>",
        "stored API key": "<redacted>",
        "stored ChatGPT tokens": "<redacted>",
        "stored agent identity": "<redacted>",
        "stored auth mode": "<redacted>"
      },
      "remediation": null,
      "durationMs": 0
    },
    "config.load": {
      "id": "config.load",
      "category": "config",
      "status": "ok",
      "summary": "config loaded",
      "details": {
        "CODEX_HOME": "C:\\Users\\<USER>\\.codex",
        "config.toml": "C:\\Users\\<USER>\\.codex\\config.toml",
        "config.toml parse": "ok",
        "cwd": "C:\\Users\\<USER>",
        "enabled feature flags": "<redacted>",
        "feature flag overrides": "none",
        "feature flags enabled": "36",
        "log dir": "C:\\Users\\<USER>\\.codex\\log",
        "mcp servers": "1",
        "model": "gpt-5.4",
        "model provider": "openai",
        "sqlite home": "C:\\Users\\<USER>\\.codex"
      },
      "remediation": null,
      "durationMs": 0
    },
    "git.environment": {
      "id": "git.environment",
      "category": "git",
      "status": "ok",
      "summary": "git version 2.52.0.windows.1",
      "details": {
        "PATH git #1": "C:\\Program Files\\Git\\cmd\\git.exe",
        "PATH git entries": "1",
        "git build options": "git version 2.52.0.windows.1; cpu: x86_64; built from commit: 2912d8e9b8253723974b7baf1c890273b1a1c5bd; sizeof-long: 4; sizeof-size_t: 8; shell-path: D:/git-sdk-64-build-installers/usr/bin/sh; rust: disabled; feature: fsmonitor--daemon; libcurl: 8.17.0; OpenSSL: OpenSSL 3.5.4 30 Sep 2025; zlib: 1.3.1; SHA-1: SHA1_DC; SHA-256: SHA256_BLK; default-ref-format: files; default-hash: sha1",
        "git exec path": "C:/Program Files/Git/mingw64/libexec/git-core",
        "git version": "git version 2.52.0.windows.1",
        "repo detected": "false",
        "selected git": "C:\\Program Files\\Git\\cmd\\git.exe"
      },
      "remediation": null,
      "durationMs": 121
    },
    "installation": {
      "id": "installation",
      "category": "install",
      "status": "ok",
      "summary": "installation looks consistent",
      "details": {
        "PATH codex #1": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\codex",
        "PATH codex #2": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\codex.cmd",
        "PATH codex entries": "2",
        "current executable": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\bin\\codex.exe",
        "install context": "npm (package C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc, bin C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\bin, resources C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex-resources, path C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex-path)",
        "managed by bun": "false",
        "managed by npm": "true",
        "managed by pnpm": "false",
        "managed package root": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex",
        "npm update target": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex"
      },
      "remediation": null,
      "durationMs": 544
    },
    "mcp.config": {
      "id": "mcp.config",
      "category": "mcp",
      "status": "ok",
      "summary": "MCP configuration is locally consistent",
      "details": {
        "configured servers": "1",
        "disabled servers": "0",
        "stdio servers": "1"
      },
      "remediation": null,
      "durationMs": 0
    },
    "network.env": {
      "id": "network.env",
      "category": "network",
      "status": "ok",
      "summary": "network-related environment looks readable",
      "details": {
        "proxy env vars": "none"
      },
      "remediation": null,
      "durationMs": 0
    },
    "network.provider_reachability": {
      "id": "network.provider_reachability",
      "category": "reachability",
      "status": "ok",
      "summary": "active provider endpoints are reachable over HTTP",
      "details": {
        "openai API base URL": "https://api.openai.com/v1 reachable (HTTP 404)",
        "openai API route probe": "https://api.openai.com/v1/<redacted> route exists (HTTP 401)",
        "reachability mode": "<redacted>"
      },
      "remediation": null,
      "durationMs": 420
    },
    "network.websocket_reachability": {
      "id": "network.websocket_reachability",
      "category": "websocket",
      "status": "ok",
      "summary": "Responses WebSocket handshake succeeded",
      "details": {
        "DNS": "2 IPv4, 0 IPv6, first IPv4",
        "auth mode": "<redacted>",
        "connect timeout": "15000 ms",
        "endpoint": "wss://chatgpt.com/backend-api/<redacted>",
        "handshake result": "HTTP 101 Switching Protocols",
        "model provider": "openai",
        "models etag present": "true",
        "provider name": "OpenAI",
        "proxy env vars": "none",
        "reasoning header": "false",
        "server model present": "false",
        "supports websockets": "true",
        "wire API": "responses"
      },
      "remediation": null,
      "durationMs": 582
    },
    "runtime.provenance": {
      "id": "runtime.provenance",
      "category": "runtime",
      "status": "ok",
      "summary": "running npm on windows-x86_64",
      "details": {
        "commit": "unknown",
        "current executable": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\bin\\codex.exe",
        "install method": "npm (package C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc, bin C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\bin, resources C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex-resources, path C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex-path)",
        "platform": "windows-x86_64",
        "version": "0.145.0"
      },
      "remediation": null,
      "durationMs": 0
    },
    "runtime.search": {
      "id": "runtime.search",
      "category": "search",
      "status": "ok",
      "summary": "search is OK (bundled)",
      "details": {
        "search command": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex\\node_modules\\@openai\\codex-win32-x64\\vendor\\x86_64-pc-windows-msvc\\codex-path\\rg.exe",
        "search command readiness": "file exists",
        "search provider": "bundled"
      },
      "remediation": null,
      "durationMs": 0
    },
    "sandbox.helpers": {
      "id": "sandbox.helpers",
      "category": "sandbox",
      "status": "ok",
      "summary": "sandbox configuration is readable",
      "details": {
        "approval policy": "OnRequest",
        "codex-linux-sandbox helper": "none",
        "execve wrapper helper": "none",
        "filesystem sandbox": "restricted",
        "network sandbox": "restricted"
      },
      "remediation": null,
      "durationMs": 1
    },
    "state.paths": {
      "id": "state.paths",
      "category": "state",
      "status": "ok",
      "summary": "state paths and databases are inspectable",
      "details": {
        "CODEX_HOME": "C:\\Users\\<USER>\\.codex (dir)",
        "active rollout files": "268 files, 4232473473 total bytes, 15792811 average bytes",
        "archived rollout files": "0 files, 0 total bytes, 0 average bytes",
        "goals DB": "C:\\Users\\<USER>\\.codex\\goals_1.sqlite (file)",
        "goals DB integrity": "ok",
        "log DB": "C:\\Users\\<USER>\\.codex\\logs_2.sqlite (file)",
        "log DB integrity": "ok",
        "log dir": "C:\\Users\\<USER>\\.codex\\log (dir)",
        "memories DB": "C:\\Users\\<USER>\\.codex\\memories_1.sqlite (file)",
        "memories DB integrity": "ok",
        "sqlite home": "C:\\Users\\<USER>\\.codex (dir)",
        "state DB": "C:\\Users\\<USER>\\.codex\\state_5.sqlite (file)",
        "state DB integrity": "ok",
        "thread history DB": "C:\\Users\\<USER>\\.codex\\thread_history_1.sqlite (missing)",
        "thread history DB integrity": "skipped (missing)"
      },
      "remediation": null,
      "durationMs": 1711
    },
    "state.rollout_db_parity": {
      "id": "state.rollout_db_parity",
      "category": "threads",
      "status": "warning",
      "summary": "rollout files and state DB thread inventory differ",
      "details": {
        "default model provider": "openai",
        "rollout DB active files": "267",
        "rollout DB active rows": "267",
        "rollout DB archive mismatches": "0",
        "rollout DB archived files": "0",
        "rollout DB archived rows": "0",
        "rollout DB duplicate DB paths": "0",
        "rollout DB duplicate rollout thread ids": "0",
        "rollout DB malformed file names": "0",
        "rollout DB missing active rows": "0",
        "rollout DB missing archived rows": "0",
        "rollout DB model providers": "openai=267",
        "rollout DB rows": "267",
        "rollout DB scan cap reached": "false",
        "rollout DB scan error sample": "C:\\Users\\<USER>\\.codex\\sessions\\2026\\02\\03\\rollout-2026-02-03T10-05-11-<REDACTED_THREAD_ID>.jsonl (rollout at C:\\Users\\<USER>\\.codex\\sessions\\2026\\02\\03\\rollout-2026-02-03T10-05-11-<REDACTED_THREAD_ID>.jsonl has no usable header record)",
        "rollout DB scan errors": "1",
        "rollout DB sources": "cli=264, subagent:thread_spawn=3",
        "rollout DB stale rows": "0"
      },
      "issues": [
        {
          "severity": "warning",
          "cause": "rollout scan was incomplete or found bad files",
          "measured": "1 scan errors, 0 malformed names, scan cap reached: false",
          "expected": "rollout directories are fully scannable",
          "remedy": "Check file permissions and unexpected files under CODEX_HOME sessions.",
          "fields": []
        }
      ],
      "remediation": null,
      "durationMs": 424
    },
    "system.environment": {
      "id": "system.environment",
      "category": "system",
      "status": "ok",
      "summary": "OS language it-IT",
      "details": {
        "EDITOR": "not set",
        "VISUAL": "not set",
        "os": "Windows 10.0.26200 (Windows 11 Core) [64-bit]",
        "os language": "it-IT",
        "os type": "Windows",
        "os version": "10.0.26200"
      },
      "remediation": null,
      "durationMs": 2
    },
    "terminal.env": {
      "id": "terminal.env",
      "category": "terminal",
      "status": "warning",
      "summary": "width 56 cols - output may wrap (recommended >=80)",
      "details": {
        "WT_SESSION": "present",
        "color output": "enabled",
        "console input code page": "850",
        "console output code page": "850",
        "stderr console mode": "0x00000007 (VT processing: true)",
        "stderr is terminal": "true",
        "stdin is terminal": "true",
        "stdout console mode": "0x00000007 (VT processing: true)",
        "stdout is terminal": "true",
        "terminal": "Windows Terminal",
        "terminal size": "56x30"
      },
      "issues": [
        {
          "severity": "warning",
          "cause": "width 56 cols - output may wrap (recommended >=80)",
          "measured": "56 x 30",
          "expected": ">= 80 columns",
          "remedy": "resize the window to at least 80 columns",
          "fields": [
            "terminal size"
          ]
        }
      ],
      "remediation": null,
      "durationMs": 1
    },
    "terminal.title": {
      "id": "terminal.title",
      "category": "title",
      "status": "ok",
      "summary": "terminal title default",
      "details": {
        "terminal title activity": "true",
        "terminal title items": "activity, project-name",
        "terminal title project source": "cwd",
        "terminal title project value": "<USER>",
        "terminal title source": "default"
      },
      "remediation": null,
      "durationMs": 0
    },
    "updates.status": {
      "id": "updates.status",
      "category": "updates",
      "status": "ok",
      "summary": "update configuration is locally consistent",
      "details": {
        "cached latest version": "0.145.0",
        "check for update on startup": "true",
        "last checked at": "2026-07-23T04:46:19.354953300Z",
        "latest version": "0.145.0",
        "latest version status": "current version is not older",
        "npm update target": "C:\\Users\\<USER>\\AppData\\Roaming\\npm\\node_modules\\@openai\\codex",
        "update action": "npm install -g @openai/codex",
        "version cache": "C:\\Users\\<USER>\\.codex\\version.json"
      },
      "remediation": null,
      "durationMs": 565
    }
  }
}

What issue are you seeing?

I am seeing what appears to be a serious Codex/model operational regression affecting latency, agent loops, context processing, and credit usage during long coding sessions.

The affected workload is BROMIND, a comparatively small narrative game project. Codex has recently become dramatically slower and more expensive even for limited patches, visual corrections, and read-only audits.

I compared the affected July workload against April-June sessions used for AFTERFALL, a much larger and substantially more intensive MMORPG project involving server, client, website, database, Linux migrations, builds, dungeons, maps, and cross-component dependencies.

The historical AFTERFALL workload was objectively more intensive:

• April-June: 47,484 tool calls
• April-June: 42,513 shell commands
• April-June: 6,146 build/operations signals
• April-June: 4,526 non-zero command exits

July had:

• 15,677 tool calls
• 11,290 shell commands
• 2,966 build/operations signals
• 975 non-zero exits

Despite the smaller operational workload, July consumed:

• 2,516,506,098 total tokens
• 2,406,631,040 cached input tokens
• 96.07% cached share
• 14,342 model cycles
• 175,464 average tokens per model cycle

Only about 98.42M input tokens were new. The dominant behavior was repeated processing of a very large cached context.

One affected gpt-5.4 session, created on July 17 and continued until July 23, recorded:

• 358,699,857 total tokens
• 356,208,797 input tokens
• 325,309,568 cached input tokens
• 30,899,229 estimated new input tokens
• 2,491,060 output tokens
• 1,116,234 reasoning output tokens
• 2,259 unique model cycles
• average 158,787 tokens per cycle
• context occupancy reaching 94.8%

Individual tasks in this session consumed between approximately 13.7M and 18.1M tokens, sometimes for work lasting only 40-75 minutes.

The same session also produced a 1.9 GB local JSONL file. Fifty compacted records occupied about 1.748 GB. Approximately 1.737 GB consisted of inline base64 images repeatedly serialized inside payload.replacement_history.

Forty-two distinct image messages appeared 1,026 times across compactions, with the same screenshot copied as many as 28 times.

The image duplication does not appear to correspond one-to-one with billed backend tokens, but it causes extreme local session growth and indicates problematic compaction/session handling.

The backend consumption problem appears broader: the dominant cost is repeated processing of cached context across thousands of model cycles. Compactions account directly for only a small percentage of tokens, but tasks containing compactions account for most of the expensive saturated-context phase.

There were also many timeouts, failed commands, and repeated tool loops while the context was close to its limit. These caused additional model cycles to process nearly the same large context.

Observed user impact:

• Severe slowdown compared with earlier Codex usage
• Small patches taking 40-75 minutes
• Repeated tool loops in 90-95% saturated contexts
• Frequent timeouts and failed-command recovery cycles
• Disproportionate credit consumption
• Multi-gigabyte local session files
• A single read-only comparative audit reduced the observed credit balance from 724 to 709 while it was running

This does not look explainable by project complexity alone. AFTERFALL generated substantially more real tool and shell activity and was subjectively much faster and more effective.

I suspect a regression in Codex/model operational behavior, agent orchestration, cached-context handling, or the session/compaction pipeline. Local logs cannot isolate which internal component changed behind the public model name.

What steps can reproduce the bug?

Uploaded thread: 019f6ec7-885b-7e42-a73d-23ef210d4e20

What is the expected behavior?

Codex should avoid repeatedly reprocessing nearly the entire historical context for every tool-loop step.

After compaction:

• old inline image payloads should be replaced by references, hashes, placeholders, or concise textual summaries;
• the same base64 image should not be serialized repeatedly into every new replacement_history;
• completed historical tool loops should be summarized instead of repeatedly reintroduced;
• cached-context size should remain bounded;
• small follow-up patches should not require hundreds of model cycles or tens of millions of tokens;
• failed commands and timeouts should not cause repeated near-full-context inference loops;
• users should receive a warning or automatic session rotation recommendation before context saturation becomes extremely expensive.

A long session should remain usable without growing to multiple gigabytes locally or consuming hundreds of millions of backend tokens.

Additional information

I performed three read-only local audits using streaming analysis rather than loading the JSONL files into memory.

Attached report:

CODEX_SESSION_PERFORMANCE_AUDITS_SANITIZED.md

CODEX_SESSION_PERFORMANCE_AUDITS_SANITIZED.md

Important distinction:

• The approximately 1.9 GB JSONL growth is primarily caused by duplicated inline images in compacted replacement histories.
• The backend token/credit consumption is primarily caused by repeated processing of very large cached contexts across thousands of model cycles.
• These are related but not identical defects.

The original affected JSONL is approximately 1.9 GB and may contain private project history, so I have not attached it publicly. I can provide targeted redacted records or hashes of duplicated image entries if requested by an OpenAI maintainer.

The issue appears as a regression in practical Codex behavior compared with earlier, objectively more intensive workloads. The available local data cannot determine whether the internal cause is the model snapshot, routing, agent loop, tool orchestration, session management, or context-compaction backend.