Fastino's GLiDE prints 64.81 on a self-run of Decision Index 0.2.1

George Maloney posted GLiDE on October 1, 2026, and Sahibzada Allahyar quoted him the same hour. Maloney’s text says the model ranks first on the Decision Index, 6.9 points over Jev, and 11.5 points over Jev on Knowledge and Reasoning, across all five areas.

The blog is by Sahibzada Allahyar and Mary Newhauser. The byline is September 30. The page metadata says October 1, 2026, 23:31 UTC. The number it prints is a self-run of the official Decision Index 0.2.1 scorer: GLiDE 64.81, Jev’s published cell 57.91, a gap of 6.90. It says GLiDE leads all five areas and 31 of 38 benchmarks. It also says 0.2.1 submissions are paused, so this score is not a row on the public board.

We did not re-download the September 28 file. That file still lists Jev 1.13.0 at 57.91, then Surogate Rune at 57.44, Decider chat at 57.33, and AutoJev-27B at 56.40. The blog’s homepage table matches those four cells, and adds simple-jev at 55.74.

Where the blog says the gap sits

Tools and Automation is 83.5 against Jev at 75.1. Knowledge and Reasoning is 62.9 against 51.4. Those are the two area lines the blog prints as skill points, +8.4 and +11.5. On single benchmarks it prints CLadder at 88.7% against 72.6%, and CRUXEval at 92.6% against 73.0%.

The docs repeat 64.81 against 57.91 and the 31 of 38. The call is POST https://api.fastino.ai/v1/systemone. The example model id is fastino/glide. A catalog snippet on the same site writes fastino/GLiDE. Context is 40,000 tokens. Input is $0.30 per million tokens and output is $0. A choice can hold up to 255 options. A request that does not fit is rejected, and thinking tokens count inside the same 40,000.

Opened October 8, 2026, the pricing page and the model catalog list GLiDE at $0.15 per million input tokens and $0 per million output tokens. The catalog names the model id fastino/GLiDE and a 40,000-token window. The concept page still prints 64.81 against 57.91.

Sahibzada writes that the model turns reasoning on when the fast distribution is uncertain, then folds that reasoning back into the probabilities. The pages we opened do not print the threshold for that switch.

The October 3 chart

On October 3 Sahibzada posted a bar chart titled as Decision Index 0.2.1, full suite. The bars are GLiDE 64.81, Cloudflare Clef 61.2, TypeSafe Jev 57.91, Clef-flash 57.1, and Perplexity pplx-decider 56.39. The axis starts at 56. The caption says the Cloudflare scores are self-reported, that Fastino ran Perplexity on the full suite, and that the Jev cell is the public leaderboard.

That 61.2 is not a composite on the Clef card this desk read. The card’s published latency row is a different comparison, 209.3 ms and 38.8 ms against a printed Jev cell of 524.1 ms. The 56.39 is also not the 7,210-row panel, where pplx-decider prints 85.71% against Jev at 84.51%.

In the same post he writes that GLiDE never trained on a Decision Index row. That sentence is his. We did not audit the training set. He also calls a “about 13x faster” line for Clef-flash a local-versus-API comparison. The Clef blog’s median row is the figure this desk already printed. We did not time either host.

We did not call api.fastino.ai, and we did not rerun the scorer. 64.81 stays a self-run against the board’s 57.91.

On October 7, 2026 the JevBench API board, release v1.6.1 under revision v1.7.12, lists Fastino GLiDE at composite 30.6, rank 11, with rounded axes 66, 88, 77, and 36, and $0.137 per 1,000 decisions. The capability list prints 77.0, Intelligence 66.1 and Calibration 88.0, and places the row outside the Jev-class cost cap. That line prints $0.14 per 1,000 decisions, 4.25 times the v1.5 Jev reference, and p50 0.74 s. We did not rerun v1.6.1.

Opened October 8, 2026, the same API board is revision v1.7.21 and ranks 26 hosted systems. Fastino GLiDE is composite rank 14 at 30.6. The composite row still prints $0.137. The capability card is still 77.0, outside the cost cap, and the rank line says official composite 14. We did not rerun the suite.

On October 7, 2026 the Decision Index space was edition 0.3. Its full-score table lists a GLiDE no-thinking row at 60.2, rank 2, model id decision-27b-ckpt-nothink. The button on that row says open weights are coming soon. Jev 1.13.0 is a reference at 60.1. Rank 1 is Perplexity Decider v1.1 at 62.8. The 60.2 row and the October 1 self-run of 64.81 are two readings. The 0.3 write-up is on the Decision Index page. We did not rerun edition 0.3.

Opened October 8, 2026, the same space headers 115 models, updated 2026-10-07, and the GLiDE no-thinking row is still 60.2. We did not rerun the index.