This page is read from MIDI Instrument Definitions, a Python library used in the author's other projects, in which every number is traced to a document its maker published. Its source is on GitHub.

Novation Summit MIDI implementation

The MIDI implementation of the Novation Summit: what it answers to, and the 250 controls it offers, on firmware 2.1.

At a glance

Firmware

This page describes firmware 2.1.

MIDI channel

Any one of 1 to 16.

Parts

Part A: on a MIDI channel of its own, taking notes, controls, and program changes, with 8 voices.
Part B: on a MIDI channel of its own, taking notes, controls, and program changes, with 8 voices.

Voices

1 or 16, as set on the instrument.

Note range

Not recorded.

Velocity

Not recorded.

Aftertouch

Channel pressure and polyphonic, each answered. It sends aftertouch too.

Pitch bend

12 semitones, and the range can be changed.

Program change

Received and sent, with 1024 presets.

MIDI clock

Received.

Start and stop

Not recorded.

Control change

74 controls, in the tables below, 18 of them with a fine half on a second controller number.

NRPN

Supported. 176 of the controls below have one.

System exclusive

Supported.

Controls

In the order the definition gives them, by group. Under each control is the name a program calls it by, once pymidiinstrumentdefs.load("novation/summit") has loaded them.

Voice

Oscillators

Mixer

Filter

Envelopes

LFOs

Effects

ARP

Animate

Modulation matrix

Settings

Where these facts come from

Source documents

• Summit User Guide, v1.1, for firmware v1.1, dated 12 July 2021, its user manual, read on 5 October 2026 (the file read).
• Summit and Peak User Guide Addendum Version 2.1, V1, for firmware updates 2.0 and 2.1, dated 27 July 2022, its user manual, read on 3 October 2026 (the file read).
• Summit User Guide v1.2 Italiano, 1.2, its user manual, read on 5 October 2026 (the file read).
• Summit Mode d'emploi V2 Francais, V2, its user manual, read on 5 October 2026 (the file read).
• Peak user guide, for firmware v1.2, dated 7 August 2019, its user manual, read on 3 October 2026 (the file read).
• Summit | Novation Downloads, its download page, read on 5 October 2026 (the file read).

How they were read

Summit User Guide v1.1 English, 62 pages, created 2021-07-12 by its own metadata, and the "Summit and Peak User Guide Addendum Version 2.1" (addendum p. 1), 18 pages, created 2022-07-27. Novation publishes no programmer's reference and no separate MIDI implementation for this instrument, unlike its Circuit, so these two documents are the whole of the map it publishes for the Summit - which the saved downloads page establishes as what was offered on the day.

BOTH END IN A MIDI PARAMETERS LIST and the two together are this definition, all but sixty-one of its numbers: 229 rows over the guide's pp. 47-49 and 19 more on the addendum's p. 18, plus three NRPNs the addendum publishes in a bug-fix bullet and in no table at all. 249 controls come out of 251 rows: one row carries NA:NA and one is a rename rather than an addition. Sixty of the 249 are rows this guide leaves without a control number, and their numbers are taken from the Peak's manual, as set out below. The 250th control is the high-pass noise filter, which this guide documents in its body and neither of this instrument's lists has a row for; its number is the Peak manual's too.

THE TABLE IS A DRAWN GRID AND THAT IS HOW IT WAS READ. Novation's publisher gives every cell a rule above and below it, each rule a separate thin rectangle whose own x extent is one column's width, so the page itself states where its five columns begin and end. That is not a convenience here: cells wrap onto two and three lines, and every page prints two of these tables side by side. Notes/4599_extract_summit.py reads the grid.

TWO THINGS ABOUT THIS DOCUMENT THAT THE PEAK'S DID NOT NEED. The blocks have to be ordered by x rather than by the height of their rules, because on p. 47 the right-hand table's first rule sits above the left-hand table's - so reading them in the order the rules appear takes the page's second column first and interleaves the table, with every row individually correct, which is the kind of wrong that does not announce itself. And counting five columns is not enough to find the table: pp. 50-55 carry the 768 factory Patch names in a five-column grid drawn the same way, so the header row is what tells them apart.

THE SIXTY BLANK CELLS WERE CHECKED THREE WAYS BEFORE BEING CALLED BLANK. The grid finds the cells empty; a plain text extraction finds nothing between the kind and the range; and no number of the form 2:0 to 16:3 occurs anywhere in the three editions of this guide that were searched - the English 1.1, the Italian 1.2 and the French V2, three languages over fourteen months - while all three print slot 1's four numbers. So this is the maker's omission and not one edition's typesetting, as far as three editions of twenty-four can show. Notes/4599_extract_summit.py asserts it for all three.

WHAT THE OTHER EDITIONS ARE FOR, AND WHAT THEY ARE NOT. Neither is read for a number or quoted for a word. The Italian settles that the version in the file name is the guide's edition rather than the firmware, because it is labelled 1.2 and still states v1.1; its text layer was regenerated by a different tool and its reading order is lost, so nothing could be extracted from it even had that been wanted, and its parameter names are translated. The French witnesses the sixty blank cells a second time. Its first page is in German, a colophon whose address and trademark notice are in that language, where its second page carries the same colophon in French, and its contents page is headed in English.

THE SIXTY NUMBERS ARE TAKEN FROM THE PEAK'S MANUAL, AND THIS IS WHAT TAKING THEM RESTS ON. The Peak is another instrument. Its manual is cited here as sibling_manual, for these sixty numbers, for the noise filter's, and for the comparison that supports taking them. Another model's document may supply a number only where the instrument's own document says the two share an engine, and this one does: "Summit is essentially a two-part, multi-timbral hybrid instrument built around a dual implementation of Peak's synth core." (guide p. 4). Six things then line up, each checked row by row in two extractions of both documents. The 64 slot rows carry the same names in both lists, Mod Matrix 7 Depth in each (guide pp. 48-49, sibling_manual pp. 42-43). The ranges are printed the same: 0-16 for each source, 0-127 for depth, 0-36 for destination. The defaults are the same: 0, 0, 64 and 0. Mod Matrix Selection is NRPN 0:125 with a range of 0-15 in both. Slot 1 is numbered identically, 1:0 to 1:3. And slot 1 is not the only agreement: every one of the 168 rows this guide's list does number carries the number the Peak's list gives that row, 163 under the same name and 5 under a name spelt or worded a little differently, and this list has no row the Peak's lacks (guide pp. 47-49, sibling_manual pp. 40-43). The Peak's manual numbers slot n as n:0 to n:3 for every n up to 16, and that is what the sixty controls carry, each as MSB x 128 + LSB. One row is printed differently and no value differs: the Peak's last, Mod Matrix 16 Destination, gives its range as 0-36 and its default as 0, without the bracketed halves its other 63 rows and this guide's 64 have. The format has no field yet for marking a source as another model's, so this account and that source's comment are the marking.

WHAT THAT DOES NOT ESTABLISH. No document read for this file prints one of the sixty numbers for the Summit, so each is an inference from a sibling and not a reading: three of the guide's twenty-four language editions were searched, and the online guide could not be fetched. Nothing here was sent to an instrument. The two lists describe different firmware, and neither is the one this file targets: the Peak's manual its v1.2 (sibling_manual p. 4), this guide the Summit's v1.1, and this definition 2.1. The addendum for 2.0 and 2.1 covers both instruments and adds to the matrix - eleven destinations under "three new destinations" (addendum p. 5), and Noise as a source "in both the Mod Matrix and FX Mod Matrix" (addendum p. 10) - while its own MIDI parameters list has no Mod Matrix row (addendum p. 18). So nothing confirms or contradicts these numbers at 2.1. And the Summit has a matrix for each Part - "each of the two Summit synths has 16 such slots" (guide p. 38) - where no page gives a slot an address of a Part's own, so which Part's matrix a number reaches is left to the channel it arrives on, as it is for every control here.

AND THE RANGES PRINTED FOR THE MATRIX DO NOT FIT THE LISTS THEY SELECT FROM, in this guide before any addendum. Every source row gives 0-16, seventeen values, where the guide lists 23 sources (guide p. 46). Every destination row gives 0-36, thirty-seven values, where the guide lists 38 destinations (guide p. 46). The Peak's manual lists 37 (sibling_manual p. 39), and FfreqSep, the last in this guide's list, is not among them. Which value selects which destination no page says, so nothing here says the range stops short of that one in particular: only that 38 names cannot fit 37 values. So the sources and the destinations of all sixteen slots say range: unknown. Their numbers are not in doubt, and their ranges are: a row and the list it selects from cannot both be right, and no document says which the instrument follows. Until 2026-10-12 they carried the printed ranges, because a control with no range reads as 0 to 127 and the format then had no way to say that a range is not established.

THIS GUIDE'S LIST HAS THE PEAK'S ROWS IN THE PEAK'S ORDER, ALL BUT ONE, AND NO ROW OF ITS OWN. The row it lacks is Osc Common Noise HPF, NRPN 0:12 (sibling_manual p. 40), for a filter this guide's body documents under the heading High-pass noise filter: displayed as NoiseHPF, with an initial value of 0 and a "Range of adjustment: 0 to 127" (guide p. 26). That number is taken too, and it is the sixty-first. It differs in kind from the sixty. This guide prints every one of their rows and leaves one cell empty, where for the noise filter its list has no row at all. So the number alone is the Peak's: the name, the range and the default below are this guide's own, from its body, and nothing else is read from the Peak's row.

AND IT RESTS ON LESS THAN THE SIXTY DO. None of the six things that line up for them can be checked here, since this list has no row to set beside the Peak's. What there is: the shared engine this guide states, and a list that carries the Peak's number on every row it does number. Against it: the Peak's own row is a blank one, its range printed 0-0 ( to+), for a filter the Peak did not have when that manual was written, as the addendum says of the Peak, "We've added a High-pass noise filter to the Osc menu to match Summit" (addendum p. 13). And this guide's list copies every other row the Peak's manual numbers and leaves out exactly this one, which may be an omission and may be deliberate. The two ranges this list leaves blank are a different kind of fact again: see below. Until 2026-10-12 the noise filter was described here and not carried.

The two printings are not identical all the same. 17 of this list's rows are printed differently from the Peak's in a range or a default cell, six of them in a value, and no matrix range is among the six.

THE DEFAULTS REST ON ONE WITNESS AND THE GUIDE HAS A SECOND ONE THAT CANNOT BE REACHED. See the comment above the definition: 105 body blocks give an Initial value keyed by a display name scoped to a menu screen, which the MIDI list's names do not carry. Five were joined by hand and three of those refused a default. The other 83 are a witness this reading could not use, and saying so is the point - a reader who wants those 36 refusals narrowed has somewhere to start, and it is not a method this file pretends to have applied.

FOUR RANGE CELLS HOLD HALF A RANGE EACH, which is a typesetting fault rather than a contradiction, and the range is carried. Measured off the page: Mixer Osc1 and Mixer Patch Level are 20-point rows carrying both halves of their range, while Mixer Osc2, Mixer Osc3, Ring 1*2 Level and Noise Level are 11-point rows carrying one half each, in pairs - the wire numbers in the first of each pair and the bracketed display half in the second. Nothing disagrees with anything; text has left its box. All six are mixer level controls and the two complete cells read 0-255 (0 to +255), so that is what the four are given.

AND THREE ROWS SHOUT. Oscillator 1/2/3 Shape Source print their bracketed half as 0 TO +2 where all 226 other rows print to in lower case. A case-sensitive reader does not mis-read them, it fails to read them at all, computes no shift and refuses three defaults the guide states perfectly well - so the match is case-insensitive and the three rows are asserted still to shout, because a false refusal looks exactly like diligence.

TWO ROWS HAVE NO RANGE AT ALL, Effects Routing (NRPN 0:89) and Chorus Type (NRPN 0:111). Their neighbours are complete, and their own rows are single-height with nothing in the column. The guide's body gives each a set of settings and no number for any of them: seven routings, "Range of adjustment: Parallel, D->R->C, D->C->R, R->D->C, R->C->D, C->D->R, C->R->D" (guide p. 36), and three chorus programs "named 2 Tap, 4 Tap and Ensemble" (guide p. 35). Until 2026-10-12 this said no other page gave either a range. The Peak's list prints 0-6 (0 to +6) and 0-2 (0 to +2) for the same two rows (sibling_manual p. 42), seven values and three. So Effects Routing carries [0, 6] and Chorus Type [0, 2], and each rests on two things. How many values there are is this instrument's own, counted in its guide's body. That they are numbered from nought is the Peak's row, taken from a sibling on the engine this guide states, as the sixty numbers are. Neither is a range this guide prints. Until 2026-10-13 both controls were left with no range, which reads as 0 to 127.

NO NAME IS GIVEN TO A VALUE OF EITHER. No page of either instrument pairs a routing with a number. A chorus program is paired with a number in one place, the Peak's panel, and from one: "Peak has three stereo Chorus programs (simply numbered 1, 2 and 3)" (sibling_manual p. 31), a two-tap chorus, four taps and an ensemble in that order, which is the order this guide names its three in. The Peak's row numbers the same three from nought, so the panel's number is not the row's, and no page says which program a value on the wire selects.

SO Chorus Type RECORDS NO DEFAULT, AND Effects Routing RECORDS 0. The chorus's row prints a bare 2 (guide p. 48). On a row numbered from nought 2 is the third of three, and the third program if the row keeps the panel's order; on the Peak's panel 2 is the second. The Peak's list prints the same 2 (sibling_manual p. 42). So does its Init Patch table, Chorus Type 2 (sibling_manual p. 39). That table's Reverb Type is 2 beside a Reverb RevSize of 64, and the three reverb presets set RevSize to 0, 64 or 127 in turn (sibling_manual p. 31), so its reverb 2 is the panel's 2, and its chorus 2 reads as the panel's as well. The Peak's definition records no default for it, as neither definition records one for the reverb's type, whose presets both bodies number 1 to 3 (guide p. 35, sibling_manual p. 31). Until 2026-10-14 this one carried the 2, because this guide's body names its three programs and numbers none, so nothing here contradicted it. The routing's column reads 0 (0) (guide p. 48), and this guide's row prints no range to check it by. The Peak's row for the same NRPN prints 0-6 (0 to +6) and the same 0 (0) (sibling_manual p. 42), where the display is the wire value, so on the range this file takes from it the two halves agree. Its body gives "Initial value: Parallel" (guide p. 36), the first routing its range of adjustment lists, which agrees with 0 if the values follow that list's order; no page says they do. Until 2026-10-14 it carried no default, because its own row prints no range to check the two halves by.

AND THE GUIDE USES THE NAME FX Routing TWICE. The FX Menu's is the seven-way setting (guide p. 36). The Settings menu's FX Routing page has a setting for each Part, sending its effects to an output, Main, AUX (guide p. 43), and the Peak's manual has no such setting. NRPN 0:89 is read as the FX Menu's: the list calls it Effects Routing between Effects Master Bypass and Delay Level (guide p. 48), and the Peak's row for the same number counts seven.

AND ONE RANGE RUNS PAST THE TABLE IT SELECTS FROM. Delay Sync Time (NRPN 0:94) is printed 0-18 (0 to +18) (guide p. 48), nineteen values, where the Delay Sync Rate table lists sixteen divisions (guide p. 45). The Arp/Clock Sync Rate table above it lists nineteen, and the three the delay's table lacks are the arpeggiator's three longest, so the delay's row reads as copied from the arpeggiator's, which is printed 0-18 too. A row and the table it selects from cannot both be right, so the control says range: unknown, as the matrix's do where their ranges run short. Until 2026-10-13 it carried [0, 18] as printed.

THE ADDENDUM COVERS THIS INSTRUMENT AND THE PEAK AND MARKS WHAT BELONGS TO WHICH, under "Update Features Exclusive to Peak" (addendum p. 12) and "Update Features Exclusive to Summit" (addendum p. 14). Four numbers that the Peak's definition left out on that evidence are this instrument's and are here. Atouch Scale, NRPN 64:3, is aftertouch scaling - "You can adjust AtouchForce between 1-10" (addendum p. 14) - under the second heading, and the Peak has no keyboard to scale. The three FM NRPNs 25:13, 25:17 and 25:21 appear in no table in either document: they are in the p. 15 bug-fix bullet "FM controls now send and respond to their NRPN messages" (addendum p. 15), in a list that talks throughout about voices 9 to 16, which an eight-voice Peak does not have, and that names this instrument outright in another of its bullets. The reverse also holds and needs no action: the two features under the Peak's own heading, envelope Delay and a high-pass noise filter, were added to the Peak "to match Summit" (addendum pp. 12-13), so this instrument already had both (guide pp. 26 and 30). Its own list numbers neither: it has no row for the high-pass noise filter and none for an envelope's Delay. So the Delay is not a control below, and the noise filter is one only on the Peak's number, as set out above. Until 2026-10-11 this said the list numbered both.

VELOCITY DOES NOTHING UNTIL A PATCH MOVES IT OFF ZERO, AND NO PAGE SAYS AN ARRIVING NOTE'S IS USED. The Amp Envelope's Velocity parameter has an initial value of 0 and "If set to zero, the volume is the same regardless of how the keys are played" (guide p. 30). The two Modulation Envelopes have the same parameter and the same initial value. So velocity can appear dead on a Summit played from its own keys, until a patch moves amp_envelope_velocity, mod_envelope_1_velocity or mod_envelope_2_velocity off zero, or gives a matrix slot Velocity as its source (guide p. 34). Where a sentence says where velocity comes from it is the keys or the keyboard, or once an arp pattern's own steps (guide p. 34); none names MIDI. The matrix's source is Keyboard velocity (guide p. 46); the guide says of aftertouch that it is taken "either from Summit's own keyboard or as external MIDI data" (guide p. 39), and has no such sentence for velocity. So velocity is not recorded. Until 2026-10-13 it said note_on: gated, with those three controls in gated_by. The Peak has the same three parameters at the same defaults and records received; which is right for that instrument is a question about that definition, not this one.

NOT RECORDED, AND WHY. The note range: the keyboard is 61 notes (guide p. 4) and no page says which 61, nor what octave shifting reaches. Transport: neither document mentions start, stop, continue or song position. Whether MIDI clock is sent: the clock source setting chooses between the internal clock and one "provided by an external device able to transmit MIDI clock" (guide p. 34), and both documents describe only the receiving side, so both would claim the half that was not established. The MIDI mode, there being no implementation chart in either document. And the clock source itself, which the MIDI parameters list gives no number for.

ONE FURTHER DOCUMENT COULD NOT BE READ. The downloads page offers "an online version of this user guide and translations" (download_page) at userguides.novationmusic.com, which answers a fetch with a Cloudflare challenge rather than the guide - twice, hours apart. Whether it is a later edition than the 1.1 PDF is therefore not established, and nothing here rests on it. The Peak's reading met the same refusal at the same domain a day earlier.