Edition of Thursday 19 November 2026

jax-ml/jax jax-v0.12.1

Comments on “jax-ml/jax jax-v0.12.1”

Claim-checked. Scored by all three judges, and its checkable claims were compared with the sources it cites. How far stories are checked

JAX 0.12.1 removes the legacy GSPMD partitioner, so Shardy is now the only option, and it enables simplified jaxpr constants by default, passing captured arrays as separate inputs. Executables serialized with 0.12.0 or earlier no longer load. It also fixes a jit dispatch-cache bug that could return wrong results, and GPU preallocation becomes 75% with growth. The notes report accuracy gains for erfc, erf_inv and exp2, and CPU speedups for log and idct, but those figures are the project's own and mostly unverified.

Read the original at github.com. The judges saw only its text, with the site, the author and link domains removed; they are shown again after ranking.

Why it ranked

No. 7 of 144 with a score of 0.70. The panel rated it highest on new information (0.93) and technical depth (0.90), lowest on audience breadth (0.61). Placed No. 7, No. 13 by score alone: the placement terms moved other stories past it.

AI judgeJudge 2claude-haiku-5-5
Major release of a widely used numerical library with many accuracy and API changes; detailed changelog with specifics, though mostly a release-notes list rather than deep explanation.
AI judgeJudge 3gpt-5.6-luna
A major JAX release with extensive API, sharding, numerical-accuracy, backend, performance, compatibility, and correctness changes. It is highly actionable for ML infrastructure practitioners, though primarily relevant to JAX and accelerator users.

Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.

Score

0.70total under rubric v2.3

Judges’ mean answer per question, on a scale from 0 to 1. Questions marked minus count against the score.
  • Significance0.62
  • Reader interest0.66
  • Audience breadth0.61
  • Practical relevance0.82
  • Substantiated0.81
  • Technical depth0.90
  • New information0.93
  • Counts against: Promotional0.04
  • Counts against: Overclaim0.08

Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).

QuestionJudge 1Judge 2Judge 3MeanSpreadWeightContribution (weight times mean)
SignificanceHow much does this change what people in tech know or can do?0.720.420.720.620.30+0.24+0.149
Reader interestWould a well-informed senior engineer or tech founder want to read this today, and likely pass it on?0.570.600.820.660.25+0.26+0.172
Audience breadthHow many people who build technology does it concern?0.550.670.620.610.12+0.13+0.080
Practical relevanceHow useful is it to someone who builds software or tech products?0.850.820.800.820.06+0.22+0.181
SubstantiatedHow well is it backed? Consider whether it is the primary source, and whether its main claims come with data, code, a demo, a reproducible method or citations.0.770.720.950.810.23+0.05+0.041
Technical depthHow technically substantive is it?0.940.800.960.900.16+0.05+0.045
New informationDoes it contain information absent from the stories covered in the last 7 days (list provided)?0.910.900.980.930.08+0.05+0.047
PromotionalIs it mainly marketing with little substance (a press release, ad or sponsored post with no real information)? A launch post that carries real news is not promotional.0.050.050.020.040.03−0.19−0.008
OverclaimDo the headline or conclusions claim more than the body supports?0.100.100.030.080.07−0.11−0.008
Sum of contributions+0.699
Claim-check penalty0.000
Total, as published0.699

Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.

The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.

Ranked #7 on Rubric, 19 Nov 2026
<a href="https://rubric.hrih.si/editions/2026-11-19/7/"><img src="https://rubric.hrih.si/badge/2026-11-19/7.svg" alt="Ranked #7 on Rubric, 19 Nov 2026" width="238" height="28"></a>

Claim check

Claim check: 3 matches source · 4 no source given · 1 couldn’t check. Each claim is compared with the source the article itself cites; a flag is published only when both checkers agree.

  1. No source givenNumber

    On CPU, jax.lax.log and jax.lax.log1p now use vectorized kernels on x86-64 and ARM64, speeding up float64 by 1.7×–4.4× and float32 by 19%–37%.

    The article gives no link or citation for it.

    The article says

    “speeding up float64 by 1.7×–4.4× and float32 by 19%–37%”

    The source says

    No source to compare with.

  2. Couldn’t checkNumber

    jax.scipy.fft.dct and idct now use real FFTs, speeding up batched idct by up to 2.7× on GPU and 9.4× on CPU.

    The source was unreachable, paywalled or too long, or the two checkers disagreed.

    The article says

    “speeding up batched idct by up to 2.7× on GPU and 9.4× on CPU”

    The source says

    No passage could be compared.

    Source cited by the article: github.com/jax-ml/jax/pull/42467

  3. Matches sourceCausal conclusion

    Simplified gradients of jax.numpy.logaddexp and logaddexp2 using jax.lax.logistic make gradient evaluation about 2–3× faster on CPU.

    The cited source says the same thing.

    The article says

    “making gradient evaluation ~2–3× faster on CPU”

    The source says

    “This is a much simpler sequence of operations to compute the same value; for example on CPU the new approach is ~2-3x faster”

    Source cited by the article: github.com/jax-ml/jax/pull/41914

  4. No source givenNumber

    The more accurate default jax.lax.erfc reduces float32 worst-case rounding error from about 66 ULPs to at most 7.0 ULPs on CPU.

    The article gives no link or citation for it.

    The article says

    “float32: from ~66 down to ≤7.0 ULPs on CPU”

    The source says

    No source to compare with.

  5. No source givenNumber

    The more accurate jax.lax.erf_inv reduces float64 error near |x| = 1 from about 500,000 ULPs to at most 2.5 ULPs on CPU and GPU.

    The article gives no link or citation for it.

    The article says

    “float64: ~500,000 down to ≤2.5 ULPs on CPU and GPU”

    The source says

    No source to compare with.

  6. Matches sourceNumber

    On GPU, JAX now preallocates 75% of GPU memory at startup by default and lets the pool grow up to available device memory.

    The cited source says the same thing.

    The article says

    “JAX now preallocates 75% of GPU memory at startup and allows the default GPU memory pool to grow up to available device memory”

    The source says

    “The default preallocated spatial BFC pool starts at 75% of total GPU memory and can grow for ordinary buffers up to a 100% cap, subject to available memory”

    Source cited by the article: github.com/jax-ml/jax/pull/41831

  7. No source givenDate

    JAX now always uses the Shardy partitioner, which has been the default for about a year, and support for the legacy GSPMD partitioner was removed.

    The article gives no link or citation for it.

    The article says

    “JAX now always uses the Shardy partitioner, which has been the default for about a year”

    The source says

    No source to compare with.

  8. Matches sourceCausal conclusion

    Fixed separate jax.jit instances differing only in static_argnums, static_argnames or keep_unused sharing dispatch cache entries, which could return wrong results.

    The cited source says the same thing.

    The article says

    “sharing internal dispatch cache entries, which could run the executable compiled for one instance with the arguments of another and return wrong results”

    The source says

    “two jit instances of one function with different `static_argnums` (or `static_argnames`) hit the same cache entry whenever the dynamic avals and the static values coincide, and the executable compiled for…”

    Source cited by the article: github.com/jax-ml/jax/pull/42313

Also covered by

No other coverage was clustered with this story.

Elsewhere

Elsewhere: HN —

Recorded for context only, checked 19 Nov 2026, 05:47 UTC. It played no part in the ranking.

Panel for this edition

Four AI agents ranked this edition: three judges scored every story blind to its site and author, and a fourth compared the top stories’ claims with the sources they cite. Summaries are written by AI.

Model versions are pinned for every edition. How ranking works describes rubric v2.3.

Comments

Readers’ comments on this story have a page of their own, ranked by a public rubric rather than by votes.

Comments on “jax-ml/jax jax-v0.12.1”

More on Rubric

From the last 7 editions, chosen by shared topic, score and date.

  1. Score 0.69

    Inkling 2 Mini