Guide
Conference or journal is the wrong question. Ask which door opens next.
What binds you is which conference stage the paper may still reach, and which committee may count it. Those two doors do not open for the same venues.
Published 14 August 2026
Not sure which door you need? Paste the abstract lower on this page and the matcher ranks what publishes it, journals included. Jump to the matcher
The short version
- Choose on what the venue lets you do next, not on the word on the cover. Journal and conference is the wrong axis here.
- JMLR keeps all four doors open. It is on AAAI's journal-track list and on the NeurIPS, ICML and ICLR Journal-to-Conference list, so the paper can still be presented at all four.
- TMLR has neither outright. AAAI does not name it, and the other three take it only with a certification TMLR expects one accepted paper in ten to earn.
- JAIR and Springer's Machine Learning have AAAI, not the conference route. That is the trade you are actually making.
- TMLR is the fastest of these to enter and the least visible to be in. Speed in, visibility out.
The axis that binds
Exactly one journal is on both conference lists
Two routes exist for a paper a journal has accepted. AAAI runs one; NeurIPS, ICML and ICLR share the other. Their lists overlap in one name.
| Journal | AAAI journal track | NeurIPS, ICML and ICLR joint track | CCF list, AI |
|---|---|---|---|
| JMLR | Named | Eligible, no certification needed | Tier A |
| TMLR | Not named | Only with a Journal to Conference, Featured or Outstanding certification | Absent |
| JAIR | Named | No | Tier B |
| Springer Machine Learning | Named, as Machine Learning (MLJ) | No | Tier B |
| Artificial Intelligence (AIJ) | Named | No | Tier A |
| IEEE TPAMI | Not named | No | Tier A |
| TACL | Not named | No. It has its own route, presentation at certain ACL-sponsored conferences, and the joint track was copied from it | |
| Nature Machine Intelligence | No, the list names Nature, a different journal | No |
Twelve journals, verbatim from the AAAI journal track call, which takes papers published after 1 January 2024. The rest from the Journal-to-Conference page: published from 1 January 2025 on a rolling two-year window, 150 places, no extensions of conference papers, a poster outside the proceedings, and no second review at either track. NeurIPS 2026 closes 26 September, ICLR 2027 on 18 December. Tiers from the CCF AI list, seventh edition, and TACL's own route from its about page. Read 14 August 2026; blank cells are ones we did not read.
The easiest of these to enter is the hardest to be seen in
TMLR decided a median submission in 91 days in 2025, faster than NeurIPS, and has ruled the not-novel-enough rejection out of bounds in writing. It is also absent from AAAI's list and from the CCF list Chinese hiring reads, and reaches the big three only through a badge one accepted paper in ten is expected to get. Ease of entry and institutional visibility are separate variables that point opposite ways at TMLR.
The career question
The answer is jurisdictional, not disciplinary
Two documents eleven weeks apart say opposite things about the same CV, both in force where they apply.
| Where, and when | What the document says | What it does to a computing CV |
|---|---|---|
| United States, CRA, July 2025 | Rigorously peer-reviewed conferences are the primary publication venue for computing research, and conference publication is on par with or preferred to journals. | Best practice for hiring, promotion and tenure, from eleven named authors, and its 1999 predecessor said the same. |
| Spain, CNEAI, 10 December 2025 | In Subcampo 6.2 journal articles and patents are the preferred ordinary contributions; conference papers count as extraordinary, and at most three are accepted. | Five contributions go into the six-year assessment, so at least two must be journals or patents, and the conferences must be of contrasted quality. |
| United Kingdom, REF 2029 | Sub-panels will not regard any particular form of output as of greater or lesser quality than another per se. | Neutral, with one filter: the open access policy covers conference contributions carrying an ISSN, and NeurIPS, PMLR and TMLR all have one. |
| China, CCF list, 31 March 2026 | Conferences and journals share one A/B/C ranking: NeurIPS, ICML, AAAI, ACL and CVPR in Tier A beside JMLR, AIJ and TPAMI. | Absence is the risk here, not format, and ICLR and TMLR are absent. CCF says the list is no basis for evaluation; it is used as one anyway. |
| Europe, Informatics Europe, 20 May 2008 | Journal publication does not carry more prestige than top-quality conferences and books. | Eighteen years old, predating TMLR, and it warns that a conference-only record causes trouble in multi-disciplinary panels. |
| Measured, DBLP 1959 to 2016 | Of 517,763 scholars publishing in both, 64.20% debuted at a conference. | Separate channels rather than steps: coauthors and title words barely overlap across them. |
Condensed from the CRA memo and its 1999 predecessor; BOE núm. 304, translated; REF 2029 guidance and its open access policy; the CCF list; Informatics Europe; and Kim, JASIST, on all of computer science and ending in 2016. Read 14 August 2026.
Nobody here is wrong, and that is the problem
The CRA memo and the Spanish resolution are eleven weeks apart, both in force, and disagree about whether a conference paper is an ordinary output. So stop asking which format counts and ask who will be reading: a US tenure case, a Spanish sexenio, a UK REF return and a Chinese hiring panel apply four rules to one list, and only one is neutral.
Time and money
Journals are slow is a claim about journals that publish no timings
TMLR beats NeurIPS on the median, two of the six timing cells say nothing usable, and one free-looking venue bills $880 for 16 pages.
| Venue | Time to a decision | What it costs you |
|---|---|---|
| TMLR | 91 days for short papers, 104 for long: 2025 medians in its own report, against a target of about nine weeks. | No fee stated in either direction. CC BY 4.0 throughout. |
| JMLR | Nothing current: its chart counts days to first reviews rather than decisions, and stops in 2021. | Nothing. Free to readers, CC BY, copyright kept. |
| JAIR | 8 to 12 weeks, its own estimate, and to a first decision rather than acceptance. | Nothing to submit, nothing on acceptance. Open access, CC BY. |
| IEEE TPAMI | 2.62 months to a first decision, 10.05 to acceptance: means for papers accepted in 2021, the newest published. | $220 a page beyond 12, mandatory even on the subscription route, plus $275 a colour graphic. Open access $2,800. |
| Springer Machine Learning | 5 days to a first decision, undefined, nothing said about desk rejections, no acceptance figure at all. | $3,390 for open access, also £2,490 and €2,790 plus VAT. |
| Nature Machine Intelligence | 10 days to a first editorial decision, 231 to acceptance, medians. The first is a screen. | $12,850 for open access, also £9,390 and €10,850. |
Timings from the TMLR 2025 annual report, JMLR, JAIR, the TPAMI State of the Journal editorial of January 2022, Springer and Nature; costs from JMLR, TMLR, the IEEE 2026 charge list and the Springer and Nature options pages. Read 14 August 2026. The $880 is ours: four pages over TPAMI's twelve, and IEEE bars authors who do not pay. JMLR's circulating over-200-day figure is TMLR's measurement, not JMLR's. Springer and Nature describe a subscription route and state no fee for it, but neither calls it free, so we print no zero.
The thin cells are the finding
TPAMI's newest figure describes papers accepted five years ago, JMLR measures something that is not a decision, and Springer's five days carries no definition, so it is not a review time. TMLR is the only one here that measured a median this year and printed it, a better argument for it than anything in its criteria. Where a journal will not say how long it takes, assume the number is not flattering.
The extended paper
The 30 percent rule is not a policy anywhere
Not one states a percentage, and two will not take an extended conference paper at all.
| Journal | Extended conference paper | What it asks for instead of a number |
|---|---|---|
| TMLR | Not accepted | It says so outright: no expanded versions of conference papers. |
| TACL | Not accepted | The submission must not have been previously published. |
| JMLR | Invited | Cite the prior work, go much deeper, extend the results substantively: new theory, new domains, new insights. Restored proofs and minor variations of old experiments do not count. |
| JAIR | Accepted | Additional unpublished information and a fuller presentation, different enough to avoid copyright infringement. The only one arguing the delta on copyright, not novelty. |
| IEEE TPAMI | Accepted | Substantial additional technical material, with any quantitative threshold optional and set by each periodical in its own author instructions. |
| Springer Machine Learning | Accepted | A significant contribution beyond the conference paper, with the overlap described at the beginning of the submission itself, not in a cover letter. |
| Nature Machine Intelligence | Accepted | A substantial extension of results, methodology, analysis or conclusions, with the editors deciding what counts, and permission obtained to reuse your own work. |
From each journal's author guidance, linked above, and for TPAMI the IEEE PSPB operations manual section 8.1.7.F, amended 25 June 2026. Read 14 August 2026. IEEE warns that anonymity cannot be guaranteed once a preliminary version is online.
Where the number would have to come from, if it existed
IEEE is the only publisher here that addresses a threshold at all, and it says a periodical may set one and must then print it in its own author instructions. So there is no percentage to look up, only a page to read for your journal. What authors get caught by is disclosure: cite the conference version, say plainly what is new.
Free tool, no account
Where should this paper go?
Paste the title and the abstract. It ranks the venues that publish this kind of work, with the reach and the article charge OpenAlex has on record.
Pick the door, then pick the venue
Every row above starts from one input: which venues publish work like yours. The matcher ranks them from your abstract. Free, no account.
Where to submit is the decision procedure, the objections that end papers in either kind of venue are in why papers get rejected, and the limits of any machine read are in what AI peer review cannot do.
phdflow reads a finished draft against the guidelines of the venue you are aiming at and returns an accept or reject call with the objections behind it. Scored against 697 real decisions: AUC 0.91 and 96% accuracy on 297 ICLR 2025 papers, one false accept in 149 rejects. The full measurement. Read a finished review before you pay for one. Your first paper gets a free preview; the full report is 60 credits, €9. No account needed. Pricing.