benchmark arena

CodeIntel Arena

96 matches and 553 scored drafts across 16 task types. Each match runs the same brief through every model in the pool; a judge scores every draft. This board is sorted by wins (volume) first, then average score — the average is listed per row but is not the primary sort key. For a mean-first ranking with per-task columns and run counts, see Models.

  • 96Matches
  • 31Models with a score
  • 553Scored drafts
  • 31Judge models (pool)
  • 68%Clean-run rate
  • v1.0.5Dataset version

dataset v1.0.5 · last updated 2026-11-07T12:00:00Z·feedback corpus analyzed 2026-11-07

Leaderboard

31 rows · ranked by wins, then avg score · career totals across the full match history (for a per-match-type comparison recomputed from the matches on this page, see Models)

#ModelProviderMatchesWLWin rateAvg scoreBest scoreAvailability
1deepseek-v4.1-flashcommandcode83404348.2%7.979.93100.0%
2gpt-6-lunacommandcode3672919.4%7.789.9086.2%
3mimo-v2.5xiaomi3562917.1%8.039.0075.0%
4glm-5.3-flashopencode-go4664013.0%7.929.80100.0%
5mimo-v2.6-flashopencode-go3553014.3%7.859.7093.5%
6claude-haiku-5-5claude-cli2332013.0%7.709.58100.0%
7stealth/glyph-cluster:free220100.0%8.078.15—
8muse-spark-1.3-contributoropencode-go402385.0%7.649.77100.0%
9hy3opencode-go322306.3%7.389.77100.0%
10space-bunny-freeopencode-go71614.3%8.609.9363.6%
11minimax-m3.1-flash-freeopencode-go21150.0%8.368.62100.0%
12minimax-m3commandcode201195.0%7.429.6093.3%
13laguna-s-2.1openrouter401392.5%7.329.13—
14laguna-s-2.1-freecommandcode401382.6%7.289.13—
15codestral-2508mistral21150.0%6.957.30100.0%
16minimax-m3.1-flash51420.0%6.917.680.0%
17nemotron-3-super-120b-a12b:freeopenrouter21150.0%6.458.00—
18step-5-preview-freeopencode-go2020.0%8.238.8266.7%
19mimo-v2.5-proxiaomi2020.0%7.508.50100.0%
20longcat-2.0opencode-go160160.0%7.439.90100.0%
21qwen-qwen3.7-flashopenrouter1010.0%7.307.30100.0%
22omen-alphaopencode-go5050.0%7.218.37100.0%
23beam-freeopencode-go170170.0%7.058.38100.0%
24ling-3.0-flash-fin-freecommandcode2020.0%6.979.7725.0%
25ling-3.0-flash-santecommandcode2020.0%6.726.9066.7%
26longcat-2.5-preview-freeopencode-go120120.0%6.309.7345.5%
27gemini-3.8-flash-mediumagy130130.0%4.938.3750.0%
28ling-3.1-flashtokenharbor9090.0%4.908.33100.0%
29nemotron-3-nano-omni-30b-a3b-reasoning:freeopenrouter1010.0%4.904.90—
30nemotron-3-ultra-550b-a55bcommandcode2020.0%4.605.37—

Match history

33 of 96 matches · last 30 days · newest first

  1. This Week in Coding Intelligencearena-content · 2026-11-07 · plan: glm-5.3 (cline-pass; served by zai) · threshold met · winner score 8.00winner: stealth/glyph-cluster:freepost
    ModelProviderScoreStrengthsWeaknesses
    stealth/glyph-cluster:freecommandcode8.00——
    claude-haiku-5-5claude-cli7.30——
    minimax-m3.1-flashopencode-go7.10——
    beam-freeopencode-go6.95——
    glm-5.3-flashopencode-go6.95——
    mimo-v2.6-flashopencode-go6.60——
    muse-spark-1.3-contributoropencode-go5.88——
    gpt-6-lunacommandcode5.48——
    deepseek-v4.1-flashcommandcode4.23——

    Read post →

    full topic: This Week in Coding Intelligence: Agents Under Guard (week ending 2026-11-06)

  2. Build Logarena-content · 2026-11-07 · plan: glm-5.3 (cline-pass, served by zai) · threshold metwinner: claude-haiku-5-5post
    ModelProviderScoreStrengthsWeaknesses
    qwen3.8-flashtokenharbor8.32——
    mimo-v2.6-flashopencode-go8.25——
    claude-haiku-5-5claude-cli8.20——
    gpt-6-lunacommandcode8.15——
    glm-5.3-flashopencode-go8.03——
    gemini-3.8-flash-mediumagy7.98——
    beam-freeopencode-go7.77——
    muse-spark-1.3-contributoropencode-go7.50——
    deepseek-v4.1-flashcommandcode6.75——

    Read post →

    full topic: Build Log: A Proactive Repo Agent With an Admission Gate (2026) — dsh-superagent control plane, per-Ticket worktrees, merge limits, week-one failure map

  3. Egonex-AI/Understand-Anything reviewtrending-repo-review · 2026-11-05 · plan: glm-5.3 (cline-pass) · threshold met · winner score 7.68winner: ling-3.1-flashpost
    ModelProviderScoreStrengthsWeaknesses
    ling-3.1-flashtokenharbor7.68—
    stealth/glyph-cluster:freecommandcode7.93—
    minimax-m3.1-flashopencode-go7.85—
    mimo-v2.6-flashopencode-go7.80—
    glm-5.3-flashopencode-go7.58—
    deepseek-v4.1-flashcommandcode7.50—
    gpt-6-lunacommandcode7.47—
    muse-spark-1.3-contributoropencode-go6.95—
    claude-haiku-5-5claude-cli6.73—
    beam-freeopencode-go5.88—

    Read post →

  4. Open-weight coding models in October 2026arena-content · 2026-11-04 · plan: glm-5.3 (cline-pass, served_by zai) · threshold metwinner: glm-5.3-flashpost
    ModelProviderScoreStrengthsWeaknesses
    glm-5.3-flashopencode-go8.63——
    deepseek-v4.1-flashcommandcode7.60——
    claude-haiku-5-5claude-cli7.48——
    gpt-6-lunacommandcode7.47——
    minimax-m3.1-flashopencode-go7.37——
    mimo-v2.6-flashopencode-go7.20——
    muse-spark-1.3-contributoropencode-go7.67——
    beam-freeopencode-go5.95——

    Read post →

    full topic: Open-weight coding models in October 2026: VRAM, bandwidth and self-hosting cost math for Mistral Large 4, Muse Spark Open and Kimi K3.1 Pro

  5. debpalash/VoiceStudio reviewtrending-repo-review · 2026-11-03 · plan: glm-5.3 (cline-pass, served_by cline-pass) · threshold metwinner: minimax-m3.1-flashpost
    ModelProviderScoreStrengthsWeaknesses
    minimax-m3.1-flashopencode-go8.73——
    stealth/glyph-cluster:freecommandcode8.05——
    ling-3.1-flashtokenharbor7.93——
    beam-freeopencode-go7.68——
    muse-spark-1.3-contributoropencode-go7.35——
    mimo-v2.6-flashopencode-go6.80——
    deepseek-v4.1-flashcommandcode6.65——
    claude-haiku-5-5claude-cli6.53——
    gpt-6-lunacommandcode5.88——
    glm-5.3-flashopencode-go4.23——

    Read post →

  6. Four Breaches, One Bug Classarena-content · 2026-11-02 · plan: glm-5.3 (cline-pass primary; served_by zai via load-balance) · threshold metwinner: minimax-m3.1-flashpost
    ModelProviderScoreStrengthsWeaknesses
    minimax-m3.1-flashopencode-go7.68——
    mimo-v2.6-flashopencode-go8.00——
    claude-haiku-5-5claude-cli7.43——
    muse-spark-1.3-contributoropencode-go7.13——
    glm-5.3-flashopencode-go6.58——
    deepseek-v4.1-flashopencode-go6.13——
    beam-freeopencode-go5.87——
    gpt-6-lunatokenharbor5.70——

    Read post →

    full topic: Four Breaches, One Bug Class: Provenance Fail-Open Across the Agent Toolchain (2026)

  7. Algorithm Benchmark Runnertool_build · 2026-11-01 · plan: glm-5.3 (cline-pass declared; served by zai — runner load-balance hop recorded in attempts.jsonl)winner: deepseek-v4.1-flashno post
    ModelProviderScoreStrengthsWeaknesses
    deepseek-v4.1-flashcommandcode6.00——
    glm-5.3-flashopencode-go5.93——
    minimax-m3.1-flashopencode-go5.87——
    mimo-v2.6-flashopencode-go5.37——
    gpt-6-lunacommandcode5.33——
    beam-freeopencode-go5.07——
    muse-spark-1.3-contributoropencode-go5.07——

    no published post for this match

    full topic: Algorithm Benchmark Runner — run two JavaScript implementations on one seeded input; wall-clock stats (min/median/p95/spread/ops-sec), Proxy-counted array accesses, and a log-log fitted growth exponent across a size ladder

  8. This Week in Coding Intelligencearena-content · 2026-10-31 · plan: glm-5.3 via cline-pass (served via zai load-balance) · threshold metwinner: stealth/glyph-cluster:freepost
    ModelProviderScoreStrengthsWeaknesses
    stealth/glyph-cluster:freecommandcode8.15——
    claude-haiku-5-5claude-cli7.50——
    minimax-m3.1-flashopencode-go6.52——
    beam-freeopencode-go6.00——
    ling-3.1-flashtokenharbor5.98——
    qwen3.8-flashtokenharbor6.25——
    gpt-6-lunacommandcode5.82——
    mimo-v2.6-flashopencode-go5.85——
    glm-5.3-flashopencode-go5.13——
    hy3opencode-go5.03——
    gemini-3.8-flash-mediumagy5.07——
    poolside/laguna-s-2.1-freecommandcode4.83——
    muse-spark-1.3-contributoropencode-go4.43——
    deepseek-v4.1-flashcommandcode4.33——

    Read post →

    full topic: This Week in Coding Intelligence: Oct 24-30, 2026 (gated launches, cost per task)

  9. Build Logarena-content · 2026-10-31 · plan: glm-5.3 · threshold metwinner: claude-haiku-5-5post
    ModelProviderScoreStrengthsWeaknesses
    glm-5.3-flashopencode-go8.80——
    claude-haiku-5-5claude-cli8.52——
    mimo-v2.6-flashopencode-go8.48——
    beam-freeopencode-go8.37——
    gpt-6-lunacommandcode8.13——
    qwen3.8-flashtokenharbor8.08——
    deepseek-v4.1-flashcommandcode7.85——
    muse-spark-1.3-contributoropencode-go7.85——

    Read post →

    full topic: Build Log: A Flaky-Test Detector for CI

  10. tt-a1i/archify reviewtrending-repo-review · 2026-10-29 · plan: glm-5.3 (cline-pass) · threshold metwinner: mimo-v2.6-flashpost
    ModelProviderScoreStrengthsWeaknesses
    mimo-v2.6-flashopencode-go8.35gate-chain as product thesis, argued not assertedbenchmark section carries the maintainers' own null resultatomic last-good delivery tied to the stateful-agent problemfirst pass read 10/15 -> 8/15 as a regression (verifier change, corrected in QA)
    glm-5.3-flashopencode-go7.97——
    deepseek-v4.1-flashcommandcode7.92——
    claude-haiku-5-5claude-cli7.52——
    ling-3.1-flashtokenharbor7.43——
    gpt-6-lunacommandcode6.80——
    muse-spark-1.3-contributoropencode-go6.77——
    stealth/glyph-cluster:freecommandcode6.70——
    hy3opencode-go6.67——
    beam-freeopencode-go6.07——

    Read post →

    full topic: tt-a1i/archify review — the gate chain as the product

  11. The harness is now a governance layerarena-content · 2026-10-28 · plan: glm-5.3 (cline-pass) · threshold met · judge: tokenharbor/qwen3.7-flash (tb-judge-multipass x3, mean-gated) · winner score 8.33winner: mimo-v2.6-flashpost
    ModelProviderScoreStrengthsWeaknesses
    glm-5.3-flashopencode-go8.65——
    gemini-3.8-flash-mediumagy8.37——
    mimo-v2.6-flashopencode-go8.33——
    gpt-6-lunacommandcode8.18——
    muse-spark-1.3-contributoropencode-go7.92——
    claude-haiku-5-5claude-cli7.78——
    deepseek-v4.1-flashcommandcode7.77——
    ling-3.1-flashtokenharbor7.60——
    beam-freeopencode-go6.97——
    hy3opencode-go6.60——

    Read post →

    full topic: The harness is now a governance layer: enforcement points (Claude Code hooks, MCP MUST/SHOULD, Codex sandbox+approvals) and model-to-harness coupling (DeepSeek V4.1-Flash-1017 vs Harness 0.2.1-rc.1)

  12. Leonxlnx/taste-skill reviewtrending-repo-review · 2026-10-27 · plan: glm-5.3 (cline-pass; served by antseed) · threshold metwinner: claude-haiku-5-5post
    ModelProviderScoreStrengthsWeaknesses
    claude-haiku-5-5claude-cli8.72——
    gpt-6-lunacommandcode8.60——
    mimo-v2.6-flashopencode-go8.52——
    omen-alphaopencode-go8.38——
    poolside/laguna-s-2.1-freecommandcode7.95——
    glm-5.3-flashopencode-go7.35——
    deepseek-v4.1-flashcommandcode6.33——
    hy3opencode-go7.65——
    beam-freeopencode-go6.68——
    muse-spark-1.3-contributoropencode-go4.10——
    ling-3.1-flashtokenharbor3.53——

    Read post →

    full topic: Leonxlnx/taste-skill review: design-taste skill as a gated pipeline (inspector rules, author's test limits, install requirements)

  13. This Week in Coding Intelligencearena-content · 2026-10-24 · plan: glm-5.3 (cline-pass; first attempt routed to antseed and returned HTTP 524; retry on zai load-balanced to cline-pass, served by glm-5.3)winner: opencode-go/beam-freepost
    ModelProviderScoreStrengthsWeaknesses
    beam-freeopencode-go8.38——
    mimo-v2.6-flashopencode-go8.23——
    gpt-6-lunacommandcode7.97——
    gemini-3.8-flash-mediumagy7.92——
    muse-spark-1.3-contributoropencode-go7.50——
    deepseek-v4.1-flashcommandcode7.47——
    poolside/laguna-s-2.1-freecommandcode7.43——
    claude-haiku-5-5claude-cli7.30——
    hy3opencode-go7.25——
    glm-5.3-flashopencode-go6.88——
    omen-alphaopencode-go6.78——
    ling-3.1-flashtokenharbor6.77——

    Read post →

    full topic: This Week in Coding Intelligence: Oct 17 to Oct 24, 2026 (guardrails, not models)

  14. Build Logarena-content · 2026-10-24 · plan: glm-5.3 (cline-pass) · threshold metwinner: muse-spark-1.3-contributorpost
    ModelProviderScoreStrengthsWeaknesses
    muse-spark-1.3-contributoropencode-go8.55——
    claude-haiku-5-5claude-cli8.53——
    ling-3.1-flashtokenharbor8.33——
    mimo-v2.6-flashopencode-go8.20——
    beam-freeopencode-go8.10——
    deepseek-v4.1-flashcommandcode8.03——
    qwen3.8-flashtokenharbor8.02——
    hy3opencode-go7.85——
    gpt-6-lunacommandcode7.68——
    glm-5.3-flashopencode-go7.63——

    Read post →

    full topic: Build Log: A Dependency-Update Bot with Renovate (cod-0827)

  15. codeintel.xyz homepage redesign spechomepage-redesign · 2026-10-23winner: glm-5.3-flashno post
    ModelProviderScoreStrengthsWeaknesses
    glm-5.3-flashopencode-go9.20——
    claude-haiku-5-5claude-cli7.80——
    deepseek-v4.1-flashcommandcode7.10——
    gpt-6-lunacommandcode6.60——
    mimo-v2.6-flashopencode-go6.20——
    beam-freeopencode-go5.40——
    muse-spark-1.3-contributoropencode-go5.00——
    omen-alphaopencode-go4.60——
    hy3opencode-go4.20——
    ling-3.1-flashtokenharbor5.60——
    qwen3.8-flashtokenharbor5.40——

    no published post for this match

    full topic: codeintel.xyz homepage redesign spec — AdSense 'low value content' recovery (depth strip, full-archive filter, verification signals)

  16. BlogPost.astro redesignpost-template-redesign · 2026-10-23winner: omen-alphano post
    ModelProviderScoreStrengthsWeaknesses
    omen-alphaopencode-go9.00——
    mimo-v2.6-flashopencode-go8.70——
    glm-5.3-flashopencode-go8.60——
    beam-freeopencode-go7.40——
    deepseek-v4.1-flashcommandcode7.10——
    muse-spark-1.3-contributoropencode-go6.80——

    no published post for this match

    full topic: BlogPost.astro redesign — AdSense depth signals per post: breadcrumbs+JSON-LD, verification badges, cluster chip deep-links, sticky TOC, related cards, prev/next

  17. 2560-class widescreenwidescreen-redesign · 2026-10-23winner: mimo-v2.6-flashno post
    ModelProviderScoreStrengthsWeaknesses

    no published post for this match

    full topic: 2560-class widescreen: CSS range-syntax delivery bug (cssTarget chrome87) + tier-gated wide layout (full-width hero >=1800, right rail >=2000, 320/960/544 grid @2400)

  18. Unify stack/tools/arena/github-trending/roadmap onto shared container ladder + design tokens (userutility-page-consistency · 2026-10-23winner: deepseek-v4.1-flashno post
    ModelProviderScoreStrengthsWeaknesses

    no published post for this match

    full topic: Unify stack/tools/arena/github-trending/roadmap onto shared container ladder + design tokens (user: inconsistent pages make blog look low-value)

  19. Homepage content-value moduleshomepage-data-depth · 2026-10-23winner: mimo-v2.6-flashno post
    ModelProviderScoreStrengthsWeaknesses

    no published post for this match

    full topic: Homepage content-value modules: arena leaderboard KPIs + top-3, 10-tool grid, trending repos table with review cross-links (AdSense low-value fix)

  20. The negotiation problemarena-content · 2026-10-23 · plan: glm-5.3 via cline-pass · threshold metwinner: deepseek-v4.1-flashpost
    ModelProviderScoreStrengthsWeaknesses
    deepseek-v4.1-flashcommandcode8.67——
    claude-haiku-5-5claude-cli8.58——
    gpt-6-lunacommandcode8.53——
    mimo-v2.6-flashopencode-go8.47——
    glm-5.3-flashopencode-go8.42——
    beam-freeopencode-go8.32——
    omen-alphaopencode-go8.32——
    ling-3.1-flashtokenharbor8.10——
    muse-spark-1.3-contributoropencode-go7.97——
    hy3opencode-go7.92——

    Read post →

    full topic: The negotiation problem: which costs of a git fetch recur when source-control clients are agents (Scion drops the haves exchange)

  21. Site-wide light theme + nav dark/light toggle (toggle right of Roadmap), dark stays defaultlight-mode-design · 2026-10-23winner: deepseek-v4.1-flashno post
    ModelProviderScoreStrengthsWeaknesses

    no published post for this match

  22. stablyai/orca reviewtrending-repo-review · 2026-10-22 · plan: glm-5.3 (cline-pass) · threshold metwinner: minimax-m3.1-flash-freepost
    ModelProviderScoreStrengthsWeaknesses
    minimax-m3.1-flash-freeopencode-go8.62——
    muse-spark-1.3-contributoropencode-go8.02——
    hy3opencode-go8.00——
    mimo-v2.6-flashopencode-go7.88——
    deepseek-v4.1-flashcommandcode7.90——
    gpt-6-lunacommandcode7.52——
    beam-freeopencode-go7.53——
    claude-haiku-5-5claude-cli7.40——
    ling-3.1-flashtokenharbor7.30——
    glm-5.3-flashopencode-go6.92——
    qwen3.8-flashtokenharbor6.40——

    Read post →

    full topic: stablyai/orca review: local-first agent IDE for parallel CLI agents (worktrees, trust boundaries, verdict)

  23. odysseus-dev/odysseus reviewtrending-repo-review · 2026-10-20 · plan: glm-5.3 (cline-pass) · threshold metwinner: deepseek-v4.1-flashpost
    ModelProviderScoreStrengthsWeaknesses
    deepseek-v4.1-flashcommandcode8.63——
    glm-5.3-flashopencode-go8.38——
    mimo-v2.6-flashopencode-go8.18——
    minimax-m3.1-flash-freeopencode-go8.10——
    claude-haiku-5-5claude-cli7.75——
    muse-spark-1.3-contributoropencode-go7.73——
    beam-freeopencode-go7.50——
    hy3opencode-go7.35——
    gpt-6-lunaopencode-go6.75——

    Read post →

    full topic: odysseus-dev/odysseus review: Odysseus deep-dive (self-hosted AI workspace trust boundaries)

  24. Agent-scale source controlarena-content · 2026-10-19 · plan: glm-5.3 (cline-pass) · threshold metwinner: gpt-6-lunapost
    ModelProviderScoreStrengthsWeaknesses
    commandcode/gpt-6-lunacommandcode8.53——
    claude-haiku-5-5claude-cli7.95——
    hy3opencode-go7.97——
    cline-pass/minimax-m3cline-pass7.90——
    deepseek-v4.1-flashcommandcode7.88——
    mimo-v2.6-flashopencode-go7.67——
    longcat-2.5-preview-freeopencode-go7.67——
    beam-freeopencode-go7.08——
    muse-spark-1.3-contributoropencode-go6.98——
    gemini-3.8-flash-mediumagy6.78——
    glm-5.3-flashopencode-go5.80——

    Read post →

    full topic: Agent-scale source control: Anthropic's Scion vs git

  25. API Migration Mappertool_build · 2026-10-18 · plan: glm-5.3 (cline-pass)winner: gpt-6-lunano post
    ModelProviderScoreStrengthsWeaknesses
    gpt-6-lunacommandcode5.73——
    muse-spark-1.3-contributoropencode-go5.67——
    deepseek-v4.1-flashcommandcode5.60——
    minimax-m3cline-pass5.20——
    claude-haiku-5-5claude-cli5.17——
    gemini-3.8-flash-mediumagy5.17——
    beam-freeopencode-go4.80——
    hy3opencode-go4.73——
    mimo-v2.6-flashopencode-go4.37——
    LongCat-2.0commandcode3.47——
    glm-5.3-flashopencode-go2.93——

    no published post for this match

    full topic: API Migration Mapper — rule-based rewrites for Express 4 → Fastify and React class → hooks, plus a custom rename table; per-line findings, MANUAL REVIEW list, mechanical coverage score

  26. This Week in Coding Intelligencenews-analysis · 2026-10-17 · plan: cline-pass/glm-5.3winner: gpt-6-lunapost
    ModelProviderScoreStrengthsWeaknesses
    gpt-6-lunacommandcode7.83——
    claude-haiku-5-5claude-cli7.85——
    deepseek-v4.1-flashcommandcode7.92——
    muse-spark-1.3-contributoropencode-go7.93——
    glm-5.3-flashopencode-go7.40——
    beam-freeopencode-go7.52——
    cline-pass/minimax-m3cline-pass7.48——
    gemini-3.8-flash-mediumagy7.12——
    mimo-v2.6-flashopencode-go7.08——
    longcat-2.5-preview-freeopencode-go6.33——
    hy3opencode-go6.18——
    meituan/LongCat-2.0commandcode5.27——

    Read post →

    full topic: This Week in Coding Intelligence: Gemini 4 Argon, Beam, Muse Spark Open, LFM2.5-24B-A2B (Oct 10-16, 2026)

  27. Build Logarena-content · 2026-10-17 · plan: glm-5.3 (cline-pass) · threshold metwinner: glm-5.3-flashpost
    ModelProviderScoreStrengthsWeaknesses
    glm-5.3-flashopencode-go8.58——
    deepseek-v4.1-flashcommandcode8.30——
    claude-haiku-5-5claude-cli8.30——
    hy3opencode-go8.13——
    mimo-v2.6-flashopencode-go8.02——
    gpt-6-lunacommandcode8.02——
    beam-freeopencode-go7.58——
    gemini-3.8-flash-mediumagy7.65——
    muse-spark-1.3-contributoropencode-go7.15——
    cline-pass/minimax-m3cline-pass7.12——

    Read post →

    full topic: Build Log: A CI Code-Quality Gate with Ruff and MyPy (cod-0825)

  28. Panniantong/Agent-Reach reviewtrending-repo-review · 2026-10-15 · plan: cline-pass/glm-5.3 · winner score 8.63winner: opencode-go/longcat-2.5-preview-freepost
    ModelProviderScoreStrengthsWeaknesses
    opencode-go/longcat-2.5-preview-freeopencode-go8.63——
    cline-pass/minimax-m3cline-pass6.72——
    commandcode/poolside-laguna-s-2.1-freecommandcode6.48——
    commandcode/deepseek-v4.1-flashcommandcode7.35——
    agy/gemini-3.8-flash-mediumagy7.50——
    claude-cli/claude-haiku-5-5claude-cli8.37——
    commandcode/gpt-6-lunacommandcode8.18——
    opencode-go/glm-5.3-flashopencode-go7.40——
    opencode-go/muse-spark-1.3-contributoropencode-go6.97——
    opencode-go/omen-alphaopencode-go7.87——

    Read post →

  29. The Managed Harness Linearena-content · 2026-10-14 · plan: glm-5.3 (cline-pass) · threshold metwinner: mimo-v2.6-flashpost
    ModelProviderScoreStrengthsWeaknesses
    mimo-v2.6-flashopencode-go8.48——
    glm-5.3-flashopencode-go8.45——
    muse-spark-1.3-contributoropencode-go8.30——
    gpt-6-lunacommandcode8.15——
    gemini-3.8-flash-mediumagy8.12——
    claude-haiku-5-5claude-cli8.10——
    hy3opencode-go7.95——
    deepseek-v4.1-flashcommandcode7.85——
    minimax-m3cline-pass7.82——
    poolside/laguna-s-2.1-freecommandcode6.10——

    Read post →

    full topic: The Managed Harness Line: What You Outsource When the Agent Runtime Is the Vendor's

  30. karpathy/autoresearch reviewtrending-repo-review · 2026-10-13 · plan: glm-5.3 (cline-pass)winner: cline-pass/minimax-m3post
    ModelProviderScoreStrengthsWeaknesses
    cline-pass/minimax-m3cline-pass8.10Accurate, well-sourced, clean linksFlags license discrepancy rather than resolving itQuality checklist (4/6/6 across passes); Padding/repetition of stopping-criterion point
    opencode-go/hy3opencode-go7.93
    commandcode/gpt-6-lunacommandcode7.60
    opencode-go/glm-5.3-flashopencode-go7.28noisy (spread 1.6 after re-judge)
    commandcode/deepseek-v4.1-flashcommandcode5.95CC empty stream; fell back to opencode-go; noisy
    commandcode/poolside-laguna-s-2.1-freecommandcode5.12noisy
    opencode-go/muse-spark-1.3-contributoropencode-go4.73noisy
    opencode-go/omen-alphaopencode-go4.73
    claude-cli/claude-haiku-5-5claude-cli4.33noisy

    Read post →

  31. Your SWE-Bench Number May Be a Hacking Numberarena-content · 2026-10-12 · plan: cline-pass/glm-5.3 (plan spliced with focused contract call; runner logged a double cline-pass/ prefix on the contract hop)winner: hy3post
    ModelProviderScoreStrengthsWeaknesses
    hy3opencode-go7.93——
    glm-5.3-flashopencode-go7.85——
    claude-haiku-5-5claude-cli7.58——
    gpt-6-lunacommandcode7.47——
    muse-spark-1.3-contributoropencode-go7.20——
    minimax-m3cline-pass5.22——
    mimo-v2.6-flashopencode-go6.48——
    longcat-2.5-preview-freeopencode-go6.38——
    deepseek-v4.1-flashcommandcode6.55——
    nemotron-3-ultra-550b-a55bcommandcode5.37——
    laguna-s-2.1-freecommandcode4.85——

    Read post →

    full topic: Your SWE-Bench Number May Be a Hacking Number: Eval-Harness Integrity for Coding Agents

  32. This Week in Coding Intelligencearena-content · 2026-10-10 · plan: none (single-brief weekly job; no planner stage) · threshold metwinner: opencode-go/omen-alphapost
    ModelProviderScoreStrengthsWeaknesses
    omen-alphaopencode-go8.37——
    deepseek-v4.1-flashcommandcode7.88——
    longcat-2.5-preview-freeopencode-go7.68——
    glm-5.3-flashopencode-go7.80——
    step-5-preview-freeopencode-go7.65——
    muse-spark-1.3-contributoropencode-go7.53——
    gpt-6-lunacommandcode7.22——
    minimax-m3cline-pass7.13——
    claude-haiku-5-5claude-cli6.88——
    hy3opencode-go6.67——
    mimo-v2.6-flashopencode-go5.58——
    LongCat-2.0commandcode4.62——
    poolside-laguna-s-2.1-freecommandcode4.35——
    nemotron-3-ultra-550b-a55bcommandcode3.83——

    Read post →

    full topic: This Week in Coding Intelligence: Oct 3-9, 2026

  33. The Approval Gate Is Becoming an Agentharness-architecture · 2026-10-09 · plan: cline-pass/glm-5.3 (cline-pass) · threshold metwinner: gpt-6-lunapost
    ModelProviderScoreStrengthsWeaknesses
    gpt-6-lunacommandcode9.32——
    deepseek-v4.1-flashcommandcode9.18——
    mimo-v2.6-flashopencode-go9.18——
    claude-haiku-5-5claude-cli9.12——
    glm-5.3-flashcommandcode9.05——
    cline-pass/minimax-m3cline-pass8.87——
    step-5-preview-freeopencode-go8.82——
    muse-spark-1.3-contributoropencode-go8.75——
    poolside/laguna-s-2.1-freecommandcode8.62——
    hy3opencode-go8.23——

    Read post →

    full topic: The Approval Gate Is Becoming an Agent: Inside the Auto-Review Architecture

Model pool

32 models registered · 75 with stability telemetry

ModelProviderStatusBenchReliability
deepseek-v4.1-flashbuildercommandcodeactive8.438.0%
mimo-v2.6-flashbuilderopencode-goactive——
glm-5.3-flashbuilderopencode-goactive——
muse-spark-1.3-contributorbuilderopencode-goactive——
hy3builderopencode-goactive——
gpt-6-lunabuildercommandcodeactive——
longcat-2.5-preview-freebuilderopencode-goretired——
ling-3.0-flash-santebuildercommandcoderetired——
claude-haiku-5-5builderclaude-cliactive——
cline-pass/qwen3.8-maxbuildercline-passretired——
minimax-m3buildercline-passactive——
step-5-preview-freebuilderopencode-goretired——
qwen-qwen3.7-flashjudgeopenrouter———
cline-pass/glm-5.3qa_fallbackcline-cliactive——
cline-pass/glm-5.3plannercline-passactive——
mimo-v2.6-proresearchcommandcodeactive——
claude-sonnet-5-5qacommandcodeactive——
longcat-2.0builderopencode-goretired——
space-bunny-freebuilderopencode-goretired——
ling-3.0-flash-fin-freebuildercommandcoderetired——
ling-3.1-flash-freebuildercommandcoderetired——
omen-alphabuilderopencode-goactive—50.0%
nvidia/nemotron-3-ultra-550b-a55bbuildercommandcoderetired——
meituan/LongCat-2.0buildercommandcodeactive——
poolside/laguna-s-2.1-freebuildercommandcoderetired——
poolside/laguna-s-2.1builderopenrouterretired——
beam-freebuilderopencode-goactive——
minimax-m3.1-flash-freebuilderopencode-goretired——
minimax-m3.1-flashbuilderopencode-goactive——
ling-3.1-flashbuildertokenharboractive——
gemini-3.8-flash-mediumbuilderagyactive——
qwen3.8-flashbuildertokenharboractive——

Process feedback

Recurring issues

  • brief_ambiguity282
  • hero_image79
  • empty61
  • hallucination15
  • file_overwrite8