Grok 4.5 in ante is such a beast!
In our own independent tb 2.1 eval, it is both the fastest and smartest, beats both Fable 5 and GPT 5.5 in their own native Harness and the cost is even lower than GLM 5.2. 🫡
We solved this long time ago. This is a common problem in harness and both Claude code and Ante suffered this long time ago and fixed it with different solutions.
Give ante a finish line, and ante will run through it on its own~
`/goal` feature is now in stable release. As everything else in ante, fluid and smooth without clunky hook mechanism
An under-appreciated aspect of Harbor is that results are auditable and reproducible so you can make trust-less claims about your agent or model.
In the era of SWE, code was the source of truth so OSS built trust. In the era of agents, evals are the source of truth and
An under-appreciated aspect of Harbor is that results are auditable and reproducible so you can make trust-less claims about your agent or model.
In the era of SWE, code was the source of truth so OSS built trust. In the era of agents, evals are the source of truth and
We independently verified GLM-5.2 (from @Zai_org) 's terminal coding ability on terminal-bench 2.1 dataset with @harborframework.
With ante as harness, it is:
- at claude opus 4.8 level (74.6%) with simple Terminus scaffolding
- only 4% away from claude opus 4.8 in its native
all trials are conducted with same constraints (5 attempts, strict timeout and hardware limitation) as official Terminal Bench 2.1 leaderboard, with a corresponding Harbor Hub link
more details at:
New major release: we reworked the permission system along with many other Quality of Life improvements
You can now cycle through 3 permission modes:
- strict: Allow only known safe tools, Ask for unknown
- auto: Ask only known dangerous tools, Allow all rest
- yolo: Everything
We've shipped a series of major upgrades to Ante's core runtime and tools over the past few releases, and the improvements show up across the board: roughly 20–30% lower latency, +10% on our internal benchmark, and far fewer failed runs — streams reconnect mid-turn, transient API
Filed off a lot of the rough edges to make it more ergonomic!
run `ante update` to force update or it will be automatically upgraded next time you launch Tui~
Just released a big release for ante!
main upgrades:
- '/providers' to have providers centric view in addition to the model centric view of `/model`
- MCP support
- DeepSeek support improvement (some providers didn't translate the `reasoning_content` well).
Local coding models may have crossed a threshold to be viable. We compared several 30B class open weight models on Terminal-Bench 2.0 and they are close to what frontier models were around ~6 months ago.
- For consumer hardware: MoE > dense when memory is the bottleneck.
- We
What it changes:
- Privacy-first teams (legal / health / finance / defense): an option besides "no agent"
- Air-gapped on-prem CI: credible local agent loop
- Hybrid routing: local default, escalate when needed
