Antigma
@antigma_labs
building substrate for self-organizing intelligence.
16 Following 933 Followers
ante 0.2 is out! This marks our commitment of a stable protocol and you can build your own harness or assistant with our highly optimized and maintained agent core
We just completed redesigned our compact algorithm it is much reliable and intelligent now, perfect for:
- Local models with limited context windows
- Frontier models with super long running tasks (500+ steps with 90 compaction still going smooth)
Show more
We just completely reworked our compaction algorithm to fit models with limited context window.
it is much more reliable and intelligent. Perfect for
- Local models
- Frontier models with super long running uninterrupted tasks (we tested a case with 90 compactions and 500+ steps)
Show more
And our favorite: /term
It's Terminal Use: drive a persistent, interactive terminal together with ante, even other TUI-based agents.
ante will orchestrate other agents living in terminal, while you can step in anytime.
Show more
Just added a new QOL updates for ante offline mode
Qwen3.8-27B is added a preset verified local models.
We promised open weights for Qwen3.8. Now, time to meet them! 🎉
⚡ Qwen3.8-27B:
- A native multimodal dense model. With just 27B parameters, it outperforms Qwen3.7-Plus overall and shines in real-world coding & office workflows.
- 262K native context, easily extendable to 1M tokens via YaRN.
- Built for builders. Highly efficient, high-quality, and licensed under Apache 2.0.
🚀 The open weights for Qwen3.8-2.4T-A95B (Max-level) have also been released recently.
Whether you're shipping lightweight applications with Qwen3.8-27B locally or building agents with Qwen3.8-2.4T-A95B, they're yours now!
Download, deploy, and build something we haven't imagined yet. 👀👇
- Hugging Face：
- ModelScope：
Show more
Grok 4.6 is added, tested, and evaluated!
Working on deepseek v4 pro 🫡
Looks a lot like a research project we did at @antigma_labs with @yiran_zhao924 (presented on AAAI 2026 symposium)
We have a cooler name though: Heapdump 😏. More to come~
One harness to run them all,
one harness to find them,
one harness to bring them all —
and in the darkness, unbind them.
ante 0.2 is coming 🧵
The Antigma for the Enigma
I was able to comment. The truth about Pi’s article actually just said thin prompt (your context) and thick harness
Harness is about capturing the invariant structural mechanism, the yang for yin(model)，the Apollo for the
Dionysus, the *Antigma* for the Enigma
Show more
I was able to comment. The truth about Pi’s article actually just said thin prompt (your context) and thick harness
Harness is about capturing the invariant structural mechanism, the yang for yin(model)，the Apollo for the
Dionysus, the *Antigma* for the Enigma
Show more
Interesting
Ante harness from @antigma_labs puts V4-Flash above Grok-4.5 in perf. 0731 gets to 82.7 on TerminalBench 2.1, which is the exact number DeepSeek cites for DeepSeek Harness. AA: 79%. Vals: 67% (though for them, it was 28x cheaper than ≈68% Grok).
Show more
We independently verified and reproduced deepseek-v4-flash-0731 eval on terminal bench with latest release of ante.
The accuracy is almost the same 82.7% with their unpublished harness! It is currently sitting on top of our leaderboard after adjusting grok 4.5 reward hacking.
Show more
@jxnlco Haha same here, this is my favourite mic to talk to my agent
I asked ante to play around with muse spark with the recently shipped new feature `/term` and it has some interesting findings
@OpenRouter @antigma_labs There are a few things that make ante great for CI:
- Single binary, fast and simple
- Ranks #1# on many evals
- Sub-agents for parallel tasks
- Many providers and models
Ante now renders LaTex natively in TUI as a celebration for the 2026 Fields Medal recipients!
Congratulations to the 2026 Fields Medal recipients! Your groundbreaking discoveries are expanding the frontiers of mathematics and inspiring generations of mathematicians. The American Mathematical Society is honored to celebrate your achievements at #ICM2026# in Philadelphia.
Show more
We moved local ocr/pdf processing into stable release; Even though most models can figure out how to process a pdf (via bash tool), having them processed locally still significantly improve the token efficiency and save llm calls and it is proved consistently in our evals.
Most importantly, ante will work even better with smaller local models 😘
Show more
gpt 5.6 in ante is tested, evaluated and shipped (with reworked providers and models picker)
The roadmap for Antigma since Day 1
