WayToClawEarn
High impactHacker News

Kimi K2.6 counterattacks in programming competition: open source model beats Claude Opus 4.7 and GPT-5.5

Moonshot AI's open source model Kimi K2.6 has won the AICC programming competition in the past two rounds, beating Claude Opus 4.7, GPT-5.5 and Gemini Pro 3.1 with programming execution results. This is the first time that a Chinese open source model has defeated all mainstream closed source models in a code competition.

WayToClawEarn EditorialPublished May 3, 2026Updated Aug 8, 2026

Editorial review of public sources · AI-assisted drafting. How we work · Original source

Core conclusion

Moonshot AI's open weight model Kimi K2.6 won two consecutive rounds in the recent AICC (AI Programming Competition), taking first place in both the D12 Word Gem Puzzle and D13 HexQuerQues challenges, beating Claude Opus 4.7, GPT-5.5 and Gemini Pro 3.1. Although the total number of medals Claude still ranks first with 8 gold and 9 medals, Kimi K2.6, as a latecomer to the open source model, won 2 gold, 1 silver and 1 bronze in only 5 rounds of competition, ** proving that the programming capabilities of the open source model are no longer weaker than the closed source head model**.

Key Points

  • Time of incident: 2026-04-30 to 2026-05-03 (AICC Competition D12-D13) -Affected objects: AI programming developers, automated workflows that rely on closed-source models, and content production teams
  • Core changes: For the first time, the open source model surpasses all closed source flagship models in programming competitions, with an API cost that is only a fraction of Claude/GPT

Background and trigger events

AICC (AI Coding Contest) is a novel AI programming benchmark testing platform. Different from traditional code reviews (such as HumanEval, SWE-Bench), AICC allows each participating model to receive the same programming prompt, write a Python TCP client algorithm, and conduct head-to-head confrontation in a real competitive environment. Only the Python standard library is allowed (no numpy, no PIL), and the difficulty of the tasks ranges from "Scrabble" to "Maze Pathfinding".

On April 30, the community noticed that Kimi K2.6 started to participate in the battle from D9 (Towers of Annoy) after missing the first 8 rounds, and made great progress all the way:

ChallengeKimi K2.6 resultsFirst placeNumber of participating models
D9 Towers of Annoy🥈 SilverClaude Opus 4.710
D10 Knights of Hanoi🥉 Bronze MedalClaude Opus 4.710
D11 StackMaxxingDNF (Unfinished)Claude Opus 4.710
D12 Word Gem Puzzle🥇 Gold MedalKimi K2.68
D13 HexQuerQues🥇 Gold MedalKimi K2.68

Key Impact (by Dimension)

DimensionsChangeWhat it means to usRecommended actions
CostKimi K2.6 API pricing is lower than Claude/GPT closed source modelProgramming automation costs can be reduced by 50-80%Switch to open source model in non-critical tasks and reduce API expenses
Open source ecologyThe open weight model won the programming competition for the first timeThe trust of the open source community will be greatly improvedThe fallback strategy of adding open source models to the automated pipeline
Chinese marketChina’s MoE model has reached the top level in the world in the field of programmingMore Chinese optimization models are available, and Chinese programming demand response is betterAccess Kimi API to do Chinese content production tasks
Model selectionNo longer only closed source giants are availableMore "sufficient and cheap" optionsEstablish a multi-model evaluation system and allocate the optimal model according to task type

AICC Latest Medal List (Top 6)

RankingModelProviderGoldSilverBronzeTotal
1Claude Opus 4.6/4.7Anthropic8019
2Gemini Pro 3.1Google2619
3Kimi K2.6Moonshot2114
4Grok Expert 4.2/4.20xAI1405
5MiMo V2-Pro / V2.5-ProXiaomi0224
6ChatGPT GPT 5.3/5.5OpenAI0033

Interestingly, GPT-5.5 only received 3 bronze medals and zero gold and zero silver in 13 rounds of competition, reflecting that OpenAI may lag behind its competitors in the "coding competition" scenario.

Adaptation suggestions

For content producers and AI Agent automation operators, the emergence of Kimi K2.6 means the following directly executable directions:

  • Reduce API budget: Migrating daily batch tasks (content rewriting, SEO copywriting generation, data cleaning) to Kimi K2.6 can save 50-80% of API costs
  • Multi-model fallback strategy: Configure open source models as failover in automated workflows such as n8n / Claude Code / OpenClaw to prevent closed source API interruptions from causing pipeline freezes
  • Chinese content optimization: Kimi K2.6, as an open source model developed by the Chinese team, is naturally better at understanding and generating Chinese, and is suitable for Chinese content production scenarios.
  • Testing and Evaluation: Before the official switch, use consistency testing to confirm that the output quality is not lower than the current closed-source model

Task List

  • Evaluate the links in the existing automated pipeline that account for the highest API cost
  • Select 2-3 non-sensitive tasks and use Kimi K2.6 for A/B testing
  • Add the fallback strategy of open source models to the automated workflow
  • Pay attention to the community’s in-depth evaluation results of Kimi K2.6 in the Coding Agent scenario

Example: Kimi K2.6 API call

terminal
curl -sS https://api.moonshot.cn/v1/chat/completions \
  -H "Authorization: Bearer $KIMI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "kimi-k2.6",
 "messages": [{"role": "user", "content": " Python TCP , Word Gem Puzzle "}],
    "max_tokens": 4096
  }'

AI coding comparison chart

waytoclawearn.com OpenAIChatGPTClaudeDeepSeekGeminiClaude Coden8nOpenClawHermes Agent

Internal link guidance

View source →

Disclaimer: this site shares educational insights only, for inspiration and reference. No outcome guarantee; external execution and decisions are your own responsibility.