THX-01
Quadrant scores
See the full quadrantScored with the public quadrant rubric (maturity × capability; bubble = adoption). Revise as evidence lands.
- Maturity5.4/10
Apache-2.0, 322M (mmBERT-base + 15M decision head); local inference and a hosted /v1/systemone API with detailed docs. API announcement says keyless; current docs require a HAL-X key. No third-party integrations or SLA.
- Capability7/10
Typed decisions trained with RLCD (proper-score reward); post-trained in 18 languages with an Azerbaijani focus. Author numbers: 98.4% average and ECE 0.003 on its own 2,843-ticket Azerbaijani/Russian/English/Turkish set, ~10 ms per request. Not on any board.
- Adoption1/100
About 24 downloads and 4 likes. No integrations or production users found.
Vendor claims
- 98.4% average on 2,843 support tickets (Clean / Corrupted / Messy / Independent) vs Jev 1.13 97.4% at 331 ms; ECE 0.003[Vendor claim — not independently verified]
- HAL-X announced a free hosted API at /v1/systemone with no key or sign-up and a 200-decisions/minute/IP limit; current API docs require a HAL-X API key.[Vendor claim — not independently verified]
News covering this model
Read this first
All results come from HAL-X AI’s own ticket set (some rows on a 200-ticket subset). It has not been run on JevBench or the Decision Index. Strong on narrow ticket routing, untested on general decisions. UnverifiedClaim.
Hosted API (11 Oct 2026)
HAL-X announced a free hosted API at https://api.hal-x.ai/v1/systemone, compatible with the TypeSafe request shape, with a 200-decisions/minute/IP limit. The announcement says no API key or sign-up is required. The current API reference says every request needs a HAL-X API key and a THX-01 decisions grant, and documents a per-key concurrency limit instead. Treat keyless access and the announced rate limit as unverified until the operator reconciles the two sources. The documented endpoints cover choice, noul, score, number, excerpt, and citations.
Score working (11 Oct 2026; proposed update)
- Maturity = 8 × 0.30 + 6 × 0.25 + 5 × 0.25 + 1 × 0.20 = 2.40 + 1.50 + 1.25 + 0.20 = 5.35 → 5.4
- Capability = 9 × 0.30 + 7 × 0.25 + 6 × 0.25 + 5 × 0.20 = 2.70 + 1.75 + 1.50 + 1.00 = 6.95 → 7.0
- Adoption = 1 × 0.35 + 1 × 0.35 + 0 × 0.30 = 0.35 + 0.35 + 0 = 0.70 → 1
Integration 5 reflects the first-party hosted /v1/systemone endpoint; no third-party framework integration or SLA was verified.
Successor: THX-02 (4 Nov 2026)
THX-02 (Apache-2.0) is a 4.5B multimodal decision model with a 60M think adapter that can write a short thought before deciding. The card reports 65.06 on the Decision Index 0.3 public index with thinking (63.83 without) and a 50 ms median on one RTX PRO 6000. These are self-run numbers on an older edition (UnverifiedClaim), and THX-02 is not in Decision Index 0.4.1 yet. See the news. This page stays scored on THX-01; scores unchanged.
Links
- Hugging Face · API reference · Base: jhu-clsp/mmBERT-base
