Published
Updated
Nace posts Drex 1.5 at $0.04, and its chart shows 199 of 231 against Jev's 201
Nace posted on September 30 that Drex 1.5 is live, with a 128k context window, and called it first on Decision Index 0.2.1 and on JevBench. The chart on the next post prints Jev 1.13.0 at 201 of 231 and Drex 1.5 at 199 of 231. Hard tier on that chart is 82 of 111 for both. The post's price is $0.04 per million input tokens. October 1 Benchmark Heaven text has no Drex 1. On October 6, 2026 Nace posted open weights and called them Drex 1.1. The card title is Drex DLM, and that card's Decision Index 0.2 cell is 52.31 against the kit's Jev cell of 51.67.
Nace posted on September 30 that Drex 1.5 is live. The text says a 128k context window, first place on Decision Index 0.2.1 and on JevBench, sub-second latency, $0.04 per million input tokens, and 250 million free tokens. The attachment on that post is a video. The next post attaches a still titled “Drex v1.5 results.”
The still’s JevBench card is 86.2%, 199 of 231, with Jev at 201 of 231. The hard-tier card is 73.9%, 82 of 111, and Jev is 82 of 111 on the same card. The Decision Index card is 58.28, rank 1, +0.37 against Jev at 57.91. Long context, 32K to 128K, is 93.4% at a median of 2.0 seconds per request. The caption says gaps under 0.25 count as ties on the board, and that 0.37 is Drex’s lead. It also says the other rows are published board scores and the Drex rows are Nace’s runs of the same official kit. The Arts column is cut off on the right edge of the still.
The rows we can read, index then raw, breadth, knowledge, language, retrieval, and tools: Drex v1.5 58.28, 67.80, 57.33, 44.2, 62.5, 62.0, 75.3. Jev 1.13.0 57.91, 68.09, 57.08, 51.4, 62.0, 55.4, 75.1. Surogate Rune 26B-A4B v3 57.44, 67.30, 56.47, 43.4, 63.1, 63.5, 71.2. Decider chat Gemma-4-31B 57.33, 67.22, 56.20, 44.3, 60.4, 63.1, 75.6. Drex v1.3 57.04, 67.08, 55.96, 42.3, 61.9, 58.8, 76.2. AutoJev-27B 56.40, 66.89, 54.93, 40.9, 63.5, 54.9, 79.3. simple-jev, Qwen3.8-27B, 55.74, 66.26, 53.91, 36.6, 62.1, 63.2, 76.2.
The product page prints the 58.28 row as scored on September 28 with the official 0.2.1 kit, ahead on 21 of 38. That 58.28 is not in the September 28 board file. The earlier write-up of that file is on the Decision Index page. The page also says Drex 1.5 is under 10 billion parameters and the smallest in the top ten. The other top-ten sizes it prints run from 25.8B to 32.7B: Rune 25.8B, Decider 32.7B, AutoJev 27.8B, simple 27.8B. Jev’s size on that chart is listed as n/a. JPT-9B is 46.89. Context is up to 128K, from 32K on Drex 1.1.
Tokens read, a median of eight identical requests, are 70 for Drex and 367 for Jev. The page ties that pair to Drex 1.5 against Jev 1.13.0. On long documents, 8K to 32K has a median of 0.65 seconds, with the whole document at 89.5% against the first 8K at 76.5%. From 32K to 128K the median is 2.0 seconds, 93.4% against 78.0%.
The same page prints a board-game table. Drex 1.5 has 122 wins and 47 draws. Jev has 87 wins. Eight games, 256 matches, the Kaggle Game Arena harness and OpenSpiel, 16 openings, both sides. Lines of Action is 26-3-3, 85.9%. Checkers is 21-3-8, 70.3%. Chess is 9-21-2, 60.9%. Dots and Boxes is 18-0-14, 56.3%. Clobber is 17-0-15, 53.1%. Reversi is 15-1-16, 48.4%. Nine Men’s Morris is 5-19-8, 45.3%. Four in a Row is 11-0-21, 34.4%.
The FAQ says training used the official training splits of the index benchmarks plus procedural data, and that evaluation used held-out splits. A note on FP8 is about Drex 1.1, not 1.5: 2,511 questions, the same answer as bf16 on 98.1%, a mean probability move of 0.015, a max of 0.29, and accuracy 92.57% against 92.49%. An example ticket is labelled recorded 2026-09-24. Billing is 0.997 against Jev at 1.000. An urgent noul is 0.732 against 0.490. Tokens are 72 against 365.
The $0.04 price, the sub-second line, and the 250 million free tokens are the post. The page we opened did not print $0.04. On October 1 we opened Benchmark Heaven again. The page still said v1.5.4 and 106 systems. A search of that text for “Drex 1” returned no match. We did not rerun the suite.
On October 6, 2026 Nace posted that Drex 1.1 is open source. The card and the repository title the weights Drex DLM. That write-up is Drex DLM. The card’s Decision Index section is edition 0.2 and prints 52.31 against the kit’s Jev cell of 51.67. The 58.28 on this page is the hosted Drex 1.5 row on edition 0.2.1. A search of the edition 0.3 space on October 7 for Drex returned no row.
This site's reading
Editorial notes evaluating claims against primary sources, contextualizing findings alongside related implementations, and defining technical terms.
Verify
Primary post is @NaceAI, September 30, 2026, 15:29 UTC. The attachment is a video. The text says Drex 1.5 is live, 128k context window, first on Decision Index 0.2.1 and on JevBench (SOTA), absolute leader in long context decisions, latency sub-second, price $0.04 per 1M input tokens, and 250M free tokens. The link is www.nace.ai/drex.
The next post, 15:33 UTC, attaches a still titled "Drex v1.5 results." Four cards: Decision Index 0.2.1 at 58.28, rank 1, +0.37 vs Jev (57.91). Long context, 32K-128K, 93.4%, median 2.0 s per request. JevBench 86.2%, 199/231, Jev 201/231. JevBench hard tier 73.9%, 82/111, Jev 82/111.
The leaderboard on that still, index / raw / breadth / knowledge / language / retrieval / tools. Drex v1.5: 58.28 / 67.80 / 57.33 / 44.2 / 62.5 / 62.0 / 75.3. Jev 1.13.0: 57.91 / 68.09 / 57.08 / 51.4 / 62.0 / 55.4 / 75.1. Surogate Rune 26B-A4B v3: 57.44 / 67.30 / 56.47 / 43.4 / 63.1 / 63.5 / 71.2. Decider chat Gemma-4-31B: 57.33 / 67.22 / 56.20 / 44.3 / 60.4 / 63.1 / 75.6. Drex v1.3: 57.04 / 67.08 / 55.96 / 42.3 / 61.9 / 58.8 / 76.2. AutoJev-27B: 56.40 / 66.89 / 54.93 / 40.9 / 63.5 / 54.9 / 79.3. simple-jev Qwen3.8-27B: 55.74 / 66.26 / 53.91 / 36.6 / 62.1 / 63.2 / 76.2. The Arts column is cut off on the right of the still.
The caption says the index is chance-corrected, areas are weighted by the square root of their benchmark count, and Arts is fixed at 10%. It says the board treats gaps under 0.25 as ties, and that Drex v1.5 leads Jev by 0.37. It says board entries are the published board scores, and that the Drex rows are Nace's runs of the same official kit.
The product page, opened the same day, prints the 58.28 row as scored on 2026-09-28 with the official 0.2.1 kit, ahead on 21 of 38, and says 58.28 is not in the September 28 board file. It also prints under 10B parameters, and calls Drex the smallest in the top ten. Other top-ten sizes it prints run from 25.8B to 32.7B: Rune 25.8B, Decider 32.7B, AutoJev 27.8B, simple 27.8B. Jev's size on that chart is n/a. JPT-9B is 46.89. Context is up to 128K, from 32K on Drex 1.1.
Tokens read, median of eight identical requests: Drex 70, Jev 367. The page ties that pair to Drex 1.5 against Jev 1.13.0. Long documents: 8K-32K median 0.65 s, whole document 89.5% against the first 8K at 76.5%. 32K-128K median 2.0 s, 93.4% against 78.0%.
Board games on the page: Drex 1.5 has 122 wins and 47 draws, Jev 87 wins. Eight games, 256 matches, Kaggle Game Arena harness, OpenSpiel, 16 openings, both sides. Lines of Action 26-3-3, 85.9%. Checkers 21-3-8, 70.3%. Chess 9-21-2, 60.9%. Dots and Boxes 18-0-14, 56.3%. Clobber 17-0-15, 53.1%. Reversi 15-1-16, 48.4%. Nine Men's Morris 5-19-8, 45.3%. Four in a Row 11-0-21, 34.4%.
The FAQ says training used the official training splits of the index benchmarks plus procedural data, and evaluation used held-out splits. The FP8 note is about Drex 1.1: 2,511 questions, the same answer as bf16 on 98.1%, mean probability move 0.015, max 0.29, accuracy 92.57% against 92.49%. An example ticket is labelled recorded 2026-09-24: billing 0.997 against Jev 1.000, urgent noul 0.732 against 0.490, 72 tokens against 365.
The $0.04 price, the sub-second line, and the 250M free tokens are the post. The page we opened did not print $0.04. On October 1, 2026 a search of https://benchmarkheaven.com/jev-models for "Drex 1" returned no match. The page still said v1.5.4 and 106 systems. We did not rerun either board.
Compare
The September 28 Decision Index file lists Jev at 57.91 and does not contain 58.28. Nace's chart says the Drex rows are its own kit runs and the other rows are the published board. +0.37 is outside the 0.25 tie band. Matilda's self-run, 59.26, is a third number, and that card says the weights are still private.
JevBench v1.5.4 ranks Jev 1.13.0 at 72.1 on 1,624 decisions, with a sealed half of Intelligence. The 199/231 figure is 231 public items, each model through its own API, on Nace's chart. The hard-tier card is a tie at 82/111. The post's "first on JevBench" is the post's claim.
The 70 against 367 token pair was already in the September 25 posts. The product page now ties that pair to Drex 1.5 against Jev 1.13.0. Liquid's chart, 58.9 against 57.9, is a different reproduction.
On October 6, 2026 Nace posted open weights and called them Drex 1.1. The card title is Drex DLM. That card's Decision Index section is edition 0.2 and prints 52.31 against the kit's Jev cell of 51.67. The 58.28 on this page is the hosted Drex 1.5 row on edition 0.2.1.
Terms
- 199/231
- Drex 1.5 on Nace's September 30 chart, 86.2% of 231 public JevBench items. Jev 1.13.0 is 201/231, 87.0%, on the same card. Hard tier is 82/111 for both.
- $0.04
- The September 30 post's price, per million input tokens. The product page we opened did not print this rate. The post also says sub-second latency and 250M free tokens.