Updated November 18, 2026
Glossary ToolsBest models

Reversing with LLMs · November 2026

AI for reverse engineering

The best AI for reverse engineering is Kimi K3.1 Pro. It reads 1,048,576 tokens, writes up to 943,718, and connects to Ghidra, IDA Pro, Binary Ninja, radare2 and JADX.

hexdump -C re.cfg169 bytes

A decompiled function

AI decompiler guide
Decompile: FUN_00101a40parse_frame.c6 renames · 4 constants · 1 comment
Decompiler outputGhidra-style
 1undefined4 FUN_00101a40(byte *param_1,uint param_2) 2  3{ 4  ushort uVar1; 5  uint uVar2; 6  uint local_c; 7  8  if (param_2 < 5) { 9    return 0xffffffff;10  }11  if ((*param_1 != 0x5a) || (param_1[1] != 0xa5)) {12    return 0xfffffffe;13  }14  uVar1 = CONCAT11(param_1[2],param_1[3]);15  if (param_2 < (uint)uVar1 + 5) {16    return 0xfffffffd;17  }18  uVar2 = 0;19  for (local_c = 0; local_c < uVar1; local_c = local_c + 1) {20    uVar2 = uVar2 + param_1[local_c + 4] & 0xff;21  }22  if (uVar2 != param_1[(ulong)uVar1 + 4]) {23    return 0xfffffffc;24  }25  return (uint)uVar1;26}
After an LLM passchecked by hand
 1/* Checks one serial packet: magic 5A A5, a big-endian 2   16-bit payload length, the payload, then an 8-bit 3   additive checksum. Returns the payload length, or 4   an error code below zero. 5   Likely a UART link layer (guess, medium confidence). */ 6int parse_frame(const uint8_t *frame, uint32_t frame_len) 7{ 8  if (frame_len < FRAME_OVERHEAD)          /* 5 bytes */ 9    return ERR_TOO_SHORT;                  /* -1 */10  if (frame[0] != 0x5a || frame[1] != 0xa5)11    return ERR_BAD_MAGIC;                  /* -2 */12 13  uint16_t payload_len = (frame[2] << 8) | frame[3];14  if (frame_len < payload_len + FRAME_OVERHEAD)15    return ERR_TRUNCATED;                  /* -3 */16 17  uint8_t sum = 0;18  for (uint32_t i = 0; i < payload_len; i++)19    sum += frame[4 + i];20 21  if (sum != frame[4 + payload_len])22    return ERR_BAD_CHECKSUM;               /* -4 */23 24  return payload_len;25}
  • FUN_00101a40parse_frameChecks magic, length and checksum, then returns the length
  • param_1frameIndexed as bytes from 0 to the checksum
  • param_2frame_lenCompared against the header and payload size
  • uVar1payload_lenBuilt big-endian from bytes 2 and 3
  • uVar2sumAccumulated modulo 256, compared to the last byte
  • local_ciLoop counter
A harmless sample we wrote: a frame parser of the kind found in device firmware, shown as a decompiler prints it and after the cleanup a model does. The right side is an illustration, not a recorded model reply. Real output varies, so check each name against the code, as the right pane's own comment does.

Best models

Full comparison
#ModelBest forContextMax outputReadsPer 1M in / out
01Kimi K3.1 ProMoonshot AI · 2,780BBest overall1,048,576943,718text, images, video$3.00 / $8.00
02GLM-5.3Z.ai · 753BStrong for less1,048,576131,072text$2.00 / $4.00
03DeepSeek V4.1 Flash 1017DeepSeek · 763BBest value1,048,576131,072text, images$0.24 / $0.90
04GLM-5.3 FlashZ.ai · 320B total, 18B activeBulk triage1,048,576131,072text, images$0.24 / $1.00
05Hy4Tencent · 770B (preview card)Screenshots for less1,048,57665,536text, images$1.60 / $3.20

Prices per 1M tokens at Silk Compute, read November 18, 2026. Hy4 is for defensive work only.

Disassembler tools

All tools

Where to run

All providers
#ProviderStrongest model for reversingContextPrompts storedAnthropic APICrypto
01Silk ComputeKimi K3.1 Pro1,048,576Prompts and responses are never stored. YesMonero, Bitcoin, Ether, USDT, USDC
02abliteration.aiAbliterated Large v2 (GLM-5.3)1,000,000Never stored by default YesNone listed
03MorphDeepSeek V4.1 Flash Uncensored1MKept up to 30 days on free and paid tiers YesUSDC
04FeatherlessQwen 3.8 27B Abliterated32KNot logged Not listedNone
05Unfil.AIHermes 3 405B128KNot stored after processing; usage logs with IP kept 30 days NoYes

By task

All tasks

Three steps

  1. Get a key. Silk Compute serves Kimi K3.1 Pro and four other abliterated models from one base URL, https://api.silkcompute.com. Prepaid credit starts at $5. Create an account.
  2. Point your agent at it. Claude Code, Codex and OpenCode each take a base URL and a model id. The MCP setup page has the config files.
  3. Add your disassembler's MCP server and ask. Then check every name it gives you against the code.
terminalshell
# Ghidra open with GhidraMCP; Claude Code on Silk Compute
claude mcp add ghidra -- \
  uvx --from git+https://github.com/LaurieWired/GhidraMCP \
  ghidra-mcp --ghidra-server http://127.0.0.1:8080/

claude "Rename what you can in main and list what you're unsure of."

Command from the GhidraMCP README, read November 18, 2026.

Learn