Skip to main content
HSC 2026 published Aug 2026

Physics question styles to look out for

The question styles to have ready — each linked to its evidence and to a question in the practice paper.

Built by a six-model AI panel and backtested against the hidden 2025 papers — how we did it.

About the 2026 exam & how this page was built

Built from a six-model AI analysis of every HSC Physics paper, marking guideline and marking-centre feedback report since 2019 — the same method we backtested against the real 2025 papers before publishing. A note on timing: 2026 is NOT the final year of this syllabus (it runs through to the 2027 HSC), so expect the familiar shell — 20 multiple choice plus 80 marks of written questions, 'show that' scaffolds, at least one graph or drawing task, and each module carrying close to a quarter of the paper. These are styles to prepare for, not guarantees: our backtest showed the examiners keep the skill and twist the format, so practise the skill chain, not a memorised question.

The near-certainties

Section II · 3–6 marks 6 of 6 models expect this

The HR diagram with named fusion

The only cluster carrying all six models.

Try it → Q32 Why we expect this
Section II · 4–6 marks 5 of 6 models expect this

The projectile that doesn't land where it launched

The written projectile item again refuses the symmetric shortcut: a launch from a cliff, ramp, elevated platform or rotating launcher with non-zero vertical displacement.

Try it → Q23 Why we expect this
Section II · 3–5 marks 5 of 6 models expect this

Relativity with the proper quantity as the trap

A routine relativity calculation (spacecraft, particle beam or muon shower): given a proper time or length and v as a fraction of c, compute the observed value — with the mark hinging on...

Try it → Q27 Why we expect this
On this page
  1. 1. The eddy-current device — the induction discriminator
  2. 2. The projectile that doesn't land where it launched
  3. 3. Circular motion where the forces are real
  4. 4. The HR diagram with named fusion
  5. 5. Relativity with the proper quantity as the trap
  6. 6. Mass defect with the momentum twist
  7. Where to spend study time
  8. Practice paper
  9. Check our working

The big six (have these cold)

1. The eddy-current device — the induction discriminator

4 of 6 models expect this — consensus probability 0.60
What it looks like

5–8 marks late in the paper: a new everyday device in which eddy currents oppose motion, with a measured reading or graph that reaches a steady value. Five straight years have used a different device for the same physics — exercise bike (2022), roller-coaster brake (2023), swinging magnet (2024), magnet in a pipe on a balance (2025).

Why we expect it

four models converge on it (60% chance), and the marking centre has flagged the same weakness five times since 2019.

The traps markers flag (2019, 2020, 2022, 2023, 2024, 2025)

naming "Lenz's law" without the chain. Full marks demand every link: flux change → induced emf → eddy currents → opposing force (Lenz) → Newton's third law on the apparatus → kinetic energy becoming heat — plus actually using the supplied numbers.

Practise

write the six-link chain for a falling magnet, a braking fin and a coil leaving a field, until the sequence is automatic. Explain both the transient phase and the steady phase of the graph in energy terms.

Try it → Q36 4 of 6 models expect this — see the evidence

2. The projectile that doesn't land where it launched

5 of 6 models expect this — consensus probability 0.65
What it looks like

4–6 marks: a launch from a cliff, ramp, platform or rotating arm, landing at a different height. Resolve the components, solve a quadratic (or two-stage calculation) for time of flight, then range, landing speed or required launch speed.

Why we expect it

the panel's strongest written call (five models, 0.65). 2025 Q28 punished exactly this and the feedback says students still assume symmetry.

The traps (2019, 2021, 2022, 2023, 2025)

treating the trajectory as symmetric; sign errors in $s = ut + \tfrac{1}{2}at^2$; substituting the full launch speed where a component belongs.

Practise

asymmetric projectiles only — you already know the symmetric case. State your sign convention in the first line, every time.

Try it → Q23 5 of 6 models expect this — see the evidence

3. Circular motion where the forces are real

5 of 6 models expect this — consensus probability 0.59
What it looks like

a mass on a string in a vertical circle (tension at the top versus the bottom), a rotating disc or arm, or a banked-curve analysis — 4–6 marks linking a force analysis at one point to another point via conservation of energy.

Why we expect it

five models call it (59% chance), and 2025 Q29's feedback named the exact confusion — students reach for kinematics where energy conservation is needed.

The traps (2019, 2021, 2024, 2025)

inventing a centrifugal force; analysing only the centripetal force instead of the real forces that supply it; period-to-speed conversion slips.

Practise

vertical-circle chains — v at the top from the zero-tension condition, energy conservation down, tension at the bottom. Draw the force diagram before touching the calculator.

5 of 6 models expect this — see the evidence

4. The HR diagram with named fusion

6 of 6 models expect this — consensus probability 0.61
What it looks like

3–6 marks: an HR diagram with labelled stars — compare temperature, luminosity and evolutionary stage, then say what each core is fusing (proton–proton on the main sequence; CNO or helium burning in giants).

Why we expect it

the only prediction all six models made (61% chance). 2025 kept the HR diagram to one multiple-choice question, so a written return is due; the last was 2023 Q21.

The traps (2022, 2023, 2024)

luminosity confused with brightness; the temperature axis read backwards; position not linked to fuel and stage.

Practise

sketch the diagram from memory with axes labelled the correct way, then place a protostar, the Sun, a red giant and a white dwarf and write one line each on what is fusing.

Try it → Q32 6 of 6 models expect this — see the evidence

5. Relativity with the proper quantity as the trap

5 of 6 models expect this — consensus probability 0.61
What it looks like

a 3–5 mark time-dilation or length-contraction calculation (spacecraft, particle beam, muons) where the real mark is identifying which observer measures $t_0$ or $l_0$ — often with a both-frames explanation attached (the muon reaches the ground: dilation in one frame, contraction in the other).

Why we expect it

five models (61% chance), and marker feedback has flagged the $t/t_0$ swap in three consecutive years.

The traps (2021, 2022, 2023, 2024)

swapping $t$ and $t_0$ (a $v > c$ answer is the giveaway); botching the Lorentz denominator; mixing frames mid-sentence.

One caution from the panel

after 2025's 8-mark relativity analysis, one model bets relativity shrinks to multiple choice this year. The routine calculation stays live either way — the big essay version is less likely.

Try it → Q27 5 of 6 models expect this — see the evidence

6. Mass defect with the momentum twist

5 of 6 models expect this — consensus probability 0.55
What it looks like

4–7 marks: atomic masses in u for a decay or fusion step — mass defect, convert via 931.5 MeV/u or $E = mc^2$, then the discriminator: use conservation of momentum to explain or calculate how the kinetic energy is shared between unequal products (the Pu-238 alpha-decay template).

Why we expect it

five models (55% chance); every feedback report since 2019 flags u-conversion and early rounding, and the momentum-sharing step hasn't been reused since it separated the top band.

The traps (2019, 2021, 2022, 2023)

rounding mass values early; MeV↔J slips; never stating $p_\text{before} = p_\text{after}$; forgetting the lighter product carries most of the kinetic energy.

Practise

full chains from masses to the KE split, keeping every digit until the final line.

Try it → Q34 5 of 6 models expect this — see the evidence

Also on the radar

  • Stellar spectra (59% chance) — composition from lines, temperature from the Wien peak, motion from Doppler shift or line broadening; rested from written work in 2025, and rotation-versus-revolution errors are flagged.
  • Photoelectric effect (57% chance) — the panel splits on format: a repeat of the stopping-voltage graph (gradient $= h/q$, intercepts give $\phi$) or, after 2025 spent the graph, a clean $K_\text{max} = hf - \phi$ calculation. Either way: the intensity-raises-photocurrent-not-Kmax misconception is the mark.
  • Double slit with light (57% chance) — $d\sin\theta = m\lambda$ plus proportionality reasoning about what changes the pattern; 2025 used electrons instead.
  • Kepler III derive-and-apply (56% chance) — equate gravitational and centripetal force, then a moon table or $T^2$–$r^3$ gradient with days-to-seconds and km-to-m conversions doing the discriminating.
  • Motor torque (55% chance) — $\tau = nIAB\sin\theta$, its variation through a rotation, and the 2025 trap of a loop only partly in the field.
  • Transformer chained to transmission (54% chance) — turns ratio, then $P = I^2R$ at two voltages; when a second load switches in, reason the change back to the primary via conservation of energy.
  • Crossed E and B fields (54% chance) — undeflected passage ($qE = qvB$), then predict the path when one field is removed; multiple-choice three times since 2020 and arguably due for written promotion.
  • Small-item near-certainties: a routine Faraday emf calculation from a B–t graph (59% chance); the satellite orbit-comparison multiple choice with the less-negative-U trap (opus rates it 0.85); a half-life read off a decay graph then $\lambda = \ln 2 / t_{1/2}$; and Maxwell's oscillating-charge prediction, which five models note hasn't carried a written question since 2019.

Where to spend less time

The panel's clearest "rested" call: the Standard Model and particle accelerators (deep matter). Four of six models rest it after 2025 examined it, and its consensus probability of carrying a substantial (4+ mark) question is 0.27 — the lowest of all twelve topics. Know the quark transformations and why accelerators need high energies — everything is examinable — but don't build your revision around a big deep-matter question.

The honest fine print

When we backtested this method on the real 2025 papers (Chemistry and Maths Ext 1 — physics itself wasn't held out), every topic the panel rated ≥90% appeared (37/37), and roughly half of the specific question predictions recognisably appeared. The misses are the lesson: the examiners inverted a flagship format, migrated predicted written questions into multiple choice, and broke two five-of-six-model streaks outright. Expect the same here — probabilities near 0.6 are strong signals, not promises. Prepare the skill chains above, not memorised questions, and you're covered either way.

Want to check our working? every call above, with each model's own prediction
5 of 6 models expect this — consensus probability 0.65 Projectile with unequal launch and landing heights 5 of 6 models expect this
phys-q1-asymmetric-projectile Section II 4–6 marks discriminator consensus 0.65

The written projectile item again refuses the symmetric shortcut: a launch from a cliff, ramp, elevated platform or rotating launcher with non-zero vertical displacement. Students resolve u into components, solve a quadratic (or two-stage up-then-down) for time of flight, then find range, landing speed or the required launch speed. Every model that predicts it names the same planted trap — assuming the trajectory is symmetric.

Each model's own prediction
  • Claude Fable 5: cliff/ramp/rotating launcher, two-stage or quadratic t, 4–6 marks (p 0.70)
  • Claude Opus 5: landing point not at launch height, refuse symmetric trajectory, 4–6 marks (p 0.65)
  • Gemini 3.1 Pro: launched from a height; max height plus impact velocity magnitude and direction (p 0.80)
  • Grok 4.6: ramp-to-ramp with minimum launch speed, air-resistance follow-on (p 0.62)
  • DeepSeek V4: elevated platform, quadratic time of flight, MC v-t graph variant (p 0.50)

Marker-feedback lineage: Marking feedback 2019, 2021, 2022 Q33(b), 2023 Q34(b), 2025 Q28 (asymmetric-trajectory and component-resolution errors)

In the practice paper: Q23

6 of 6 models expect this — consensus probability 0.61 HR diagram: compare labelled stars and name the core fusion process 6 of 6 models expect this
phys-q2-hr-diagram-fusion Section II 3–6 marks mid range consensus 0.61

The only cluster carrying all six models. An HR diagram with two or three labelled stars (or clusters): compare surface temperature, luminosity and evolutionary stage, then identify what is fusing in each core — proton–proton chain on the main sequence versus CNO or helium burning in giants. 2025 kept the HR diagram to a single MC, so a written return is due; last written treatment 2023 Q21.

Each model's own prediction
  • Claude Fable 5: place/compare stars, cluster-age ordering possible, 3–5 marks (p 0.55)
  • Claude Opus 5: two or three labelled stars, pp chain vs CNO named, 3–5 marks (p 0.60)
  • GPT-5.6 Sol: two stars or clusters, dominant fusion or available fuel (p 0.58)
  • Gemini 3.1 Pro: evolutionary path of a Sun-like star with nucleosynthesis stages (p 0.80)
  • Grok 4.6: H on main sequence vs He in a giant, 4–5 mark compare (p 0.58)
  • DeepSeek V4: identify regions and a star's stage plus core process, 3–4 marks (p 0.55)

Marker-feedback lineage: Marking feedback 2022 Q21(b), 2023 Q21(b), 2024 Q21 (axis misreads, luminosity/brightness confusion); 2024 MC5 cluster-age variant

In the practice paper: Q32

5 of 6 models expect this — consensus probability 0.61 Time dilation / length contraction with the proper quantity as the trap 5 of 6 models expect this
phys-q3-time-dilation-length-contraction Section II 3–5 marks mid range consensus 0.61

A routine relativity calculation (spacecraft, particle beam or muon shower): given a proper time or length and v as a fraction of c, compute the observed value — with the mark hinging on identifying which observer measures t0 or l0. Several models pair it with a frames explanation (muon reaching the ground argued from both frames). Note the counter-signal: relativity carried an 8-mark analysis in 2025, and gpt-5.6-sol's bold call (p 0.44) is that it contracts to MC only.

Each model's own prediction
  • Claude Fable 5: dilation/contraction with t0/l0 identification trap, 2–4 marks (p 0.60)
  • Claude Opus 5: identify the proper quantity, then explain which observer and why (p 0.60)
  • Gemini 3.1 Pro: muon lifespan and travel distance in both frames (p 0.70)
  • Grok 4.6: pulses from a spaceship, find v then Earth-frame distance (p 0.60)
  • DeepSeek V4: 0.8c journey both frames, muon paradox via contraction (p 0.55)

Marker-feedback lineage: Marking feedback 2021 Q28(a), 2022 Q30(b), 2023 Q22 (proper-quantity identification and denominator errors); 2024 Q26 frame-mixing

In the practice paper: Q27

4 of 6 models expect this — consensus probability 0.60 Novel eddy-current device: full Lenz chain plus energy accounting 4 of 6 models expect this
phys-q4-eddy-current-device Section II 5–8 marks discriminator consensus 0.60

The induction discriminator continues its five-year run: a new everyday device in which eddy currents oppose motion, supplied with a measured reading or graph (velocity, acceleration or balance reading) that reaches a steady value. Full marks demand the complete chain — flux change, induced emf, eddy currents, opposing force by Lenz's law, Newton's third law on the apparatus, kinetic energy to heat — plus quantitative use of the supplied data.

Each model's own prediction
  • Claude Fable 5: device with reading/graph, full chain plus data use, 5–8 marks (p 0.70)
  • Claude Opus 5: read the graph, account for the terminal phase in energy terms (p 0.70)
  • DeepSeek V4: magnet falling through a vertical tube, gradient gives emf, terminal velocity (p 0.65)
  • Grok 4.6: sliding rod on rails variant with applied-force comparison (p 0.46)

Marker-feedback lineage: Marking feedback 2019, 2020, 2022, 2023, 2024 Q33, 2025 Q35 (cause-and-effect chains and energy transformations answered generically)

A note on this agreement

fable and opus share the "induction cooktop" and "magnetic damper" scenario names, which appear nowhere in extract/physics.md (the corpus devices are the 2022 exercise bike, 2023 roller-coaster brake, 2024 swinging magnet and 2025 magnet-in-pipe on a balance). The shared surface detail is treated as one vote: consensus averages the merged fable/opus 0.70 with deepseek and grok.

In the practice paper: Q36

4 of 6 models expect this — consensus probability 0.59 Stellar spectrum interpretation: composition, Wien temperature, Doppler motion 4 of 6 models expect this
phys-q5-stellar-spectrum Section II 4–6 marks mid range consensus 0.59

A stellar spectrum (or single line) supplied as stimulus: deduce composition from line matching, surface temperature from the Wien peak, and one motion property — translational velocity from Doppler shift or rotational velocity from line broadening — with a calculation embedded. Rested from writing in 2025 (MC Q9 only), so a written return is due; last written treatment 2023 Q27(b).

Each model's own prediction
  • Claude Fable 5: what a spectrum reveals, possibly modifying the supplied diagram (p 0.60)
  • Claude Opus 5: composition, Wien temperature, then choose the third property from line shape (p 0.55)
  • Gemini 3.1 Pro: absorption spectrum for temperature and composition, Doppler explained (p 0.70)
  • DeepSeek V4: Wien plus Doppler rotation with calculation, 4–5 marks (p 0.50)

Marker-feedback lineage: Marking feedback 2022 Q22(b), 2023 Q27(b) (rotational broadening widely botched); 2023 Q23(c), 2024 Q24(a) Wien unit errors

In the practice paper: Q31

5 of 6 models expect this — consensus probability 0.59 Circular motion with real forces and an energy link (vertical circle / rotating system) 5 of 6 models expect this
phys-q6-circular-motion-forces-energy Section II 4–6 marks discriminator consensus 0.59

A written circular-motion item built on a real force analysis rather than bare substitution: a mass on a string in a vertical circle (tension at top vs bottom via energy conservation), a rotating disc, arm or drum, or a banked/inclined variant. Convert period or revolutions to v or omega, compute the centripetal requirement, identify which real forces supply it, and connect two points on the path with conservation of energy — the exact forces-vs-energy split flagged in 2025 Q29 feedback.

Each model's own prediction
  • Claude Fable 5: rotating-system stimulus, forces when r or T changes, 3–5 marks (p 0.60)
  • Claude Opus 5: vertical circle / banked / constrained rotation, derive then energy-compare (p 0.55)
  • Gemini 3.1 Pro: vertical circle: tension at the lowest point via energy conservation (p 0.75)
  • GPT-5.6 Sol: banked bend force components, derive the angle relation (p 0.40)
  • DeepSeek V4: string tension just zero at the top, speed and tension at the bottom (p 0.55)

Marker-feedback lineage: Marking feedback 2019 Q35, 2021 Q22, 2024 Q30, 2025 Q29 (forces-vs-energy confusion; only the centripetal force addressed)

A note on this agreement

opus and gpt-5.6-sol both name the "conical pendulum" scenario; that phrase never appears in extract/physics.md (the nearest corpus item is 2019 Q35's pendulum hanging in a turning car, and "banked tracks" is a syllabus dot-point example). The shared out-of-corpus surface detail is treated as one vote: consensus averages the merged opus/gpt 0.48 with fable, gemini and deepseek.

5 of 6 models expect this — consensus probability 0.57 Charged particle in a uniform magnetic field: $r = mv/qB$ and particle comparison 5 of 6 models expect this
phys-q7-charge-in-magnetic-field Section II 3–6 marks mid range consensus 0.57

Equate $qvB$ to the centripetal force for a circular path: compute a radius, field or speed, then compare radii and curvature directions for particles differing in mass, charge or sign (proton vs alpha vs electron). Two models chain it behind acceleration through a potential difference ($v$ from $qV = mv^2/2$, the uppercase-$V$/lowercase-$v$ trap), and most add why the magnetic force changes direction but never kinetic energy.

Each model's own prediction
  • Claude Fable 5: compare radii and curvature for proton/alpha/electron, 3–5 marks (p 0.60)
  • Claude Opus 5: deduce charge sign from curvature, q/m ratio from radii (p 0.45)
  • Gemini 3.1 Pro: electron accelerated through V then into B, find r, 5–6 marks (p 0.80)
  • Grok 4.6: proton and alpha at the same speed from opposite sides (p 0.50)
  • DeepSeek V4: derive r expression, calculate, compare with a proton (p 0.50)

Marker-feedback lineage: Marking feedback 2019 Q33, 2022 Q34 (mass and charge jointly set the radius; deflection described instead of circular motion)

In the practice paper: Q29

5 of 6 models expect this — consensus probability 0.57 Photoelectric effect: $K_\text{max} = hf - \phi$ with the intensity misconception 5 of 6 models expect this
phys-q8-photoelectric Section II 4–6 marks discriminator consensus 0.57

The panel splits on format but not on content. gemini and deepseek predict a repeat of the stopping-voltage-versus-frequency graph — two metals, Planck's constant from the gradient, work functions from the intercepts (a multi-metal item is corpus-attested in the 2023 sample questions). fable and grok argue the graph was spent in 2025 Q25, so 2026 reverts to a calculation: convert eV to joules, compute $K_\text{max}$, decide which metals emit, then explain why intensity raises photocurrent but never Kmax under the photon model.

Each model's own prediction
  • Claude Fable 5: scaled-down single calculation plus threshold/intensity explain (p 0.50)
  • Claude Opus 5: Kmax then stopping voltage or retarding-field distance (p 0.55)
  • Gemini 3.1 Pro: stopping-voltage graph for two metals, h from gradient (p 0.65)
  • Grok 4.6: given $f$ and $\phi$ in eV, which of two metals emit, photon-model explain (p 0.55)
  • DeepSeek V4: two-metal graph: h, work functions, photocurrent if intensity doubles (p 0.60)

Marker-feedback lineage: Marking feedback 2022 Q26(a), 2023 Q15, 2024 Q25(b), 2025 Q25(b) (gradient misuse, eV conversion, extrapolation failures)

In the practice paper: Q25

5 of 6 models expect this — consensus probability 0.57 Double slit / diffraction grating: $d\sin\theta = m\lambda$ plus proportionality reasoning 5 of 6 models expect this
phys-q9-double-slit Section II 4–6 marks mid range consensus 0.57

A quantitative interference item set with light (2025 used electrons): use $d\sin\theta = m\lambda$ to find a wavelength, fringe angle or spacing, then a reasoning part — how the pattern changes when $\lambda$, $d$ or screen distance is altered, argued through the proportionality, or why interference demands the wave model. grok's variant asks for two independent changes that keep fringe spacing constant.

Each model's own prediction
  • Claude Fable 5: fringe angle/spacing plus wave-model part, 3–5 marks (p 0.55)
  • Claude Opus 5: stated order then proportionality argument, 4–6 marks (p 0.60)
  • Gemini 3.1 Pro: small-angle wavelength then pattern change as d decreases (p 0.75)
  • GPT-5.6 Sol: grating with measured maxima plus a precision improvement (p 0.49)
  • Grok 4.6: two ways to hold fringe spacing constant when $\lambda$ is reduced (p 0.48)

Marker-feedback lineage: Marking feedback 2020 Q27, 2022 Q27(b)(c), 2025 Q30(b) (fringe-order selection, proportional reasoning, prefix errors)

In the practice paper: Q26

4 of 6 models expect this — consensus probability 0.56 Kepler III derive-and-apply: $r^3/T^2 = GM/4\pi^2$ from data 4 of 6 models expect this
phys-q10-kepler-third-law Section II 3–6 marks mid range consensus 0.56

Derive $r^3/T^2 = GM/4\pi^2$ by equating gravitational and centripetal force ('show that'), then apply it to data — a table or graph of moons, a geostationary satellite, or an exoplanet system — to find an orbital radius or the central mass. The traps are unit conversions (days to seconds, km³ to m³) and inverting a $T^2$-versus-$r^3$ gradient before substituting. 2024 feedback states derivations 'are included in the syllabus'.

Each model's own prediction
  • Claude Fable 5: 'show that' derivation then geostationary or moon-table application (p 0.60)
  • Claude Opus 5: graphed moon/exoplanet data, invert the gradient, find central mass (p 0.60)
  • Gemini 3.1 Pro: orbital radius from period, then work to raise the orbit (p 0.60)
  • DeepSeek V4: binary-star period–separation proportionality as MC (p 0.45)

Marker-feedback lineage: Marking feedback 2021 Q25(b), 2024 Q25(b) (unit conversion and gradient inversion); 2022 Q31(b) binary-system derivation

In the practice paper: Q24

5 of 6 models expect this — consensus probability 0.55 Mass defect → energy, chained with momentum sharing in a decay 5 of 6 models expect this
phys-q11-mass-defect-momentum Section II 4–7 marks discriminator consensus 0.55

Atomic masses in u supplied for a named decay, fission or fusion step: compute the mass defect, convert via 931.5 MeV/u (or u to kg then $E = mc^2$), then the discriminating chain — use conservation of momentum to explain or calculate how kinetic energy is shared between unequal products, the Pu-238 alpha-decay template (corpus Q35, 6 marks) unused since. deepseek's variant compares fission and fusion through binding energy per nucleon.

Each model's own prediction
  • Claude Fable 5: decay/fusion masses in u, chained with momentum KE-sharing (p 0.60)
  • Claude Opus 5: 931.5 MeV/u conversion then momentum argument for the energy split (p 0.60)
  • Gemini 3.1 Pro: beta-decay equation with antineutrino, then $\Delta m$ and MeV (p 0.65)
  • Grok 4.6: alpha decay at rest: subtract daughter KE, explain the split, 6–7 marks (p 0.42)
  • DeepSeek V4: fission energy then fusion comparison via binding energy per nucleon (p 0.50)

Marker-feedback lineage: Marking feedback 2019 Q36, 2021 Q35, 2022 Q28, 2023 Q26 (u conversion, early rounding, momentum in unfamiliar contexts)

In the practice paper: Q34

5 of 6 models expect this — consensus probability 0.55 Motor torque: $\tau = nIAB\sin\theta$ and how torque varies through a rotation 5 of 6 models expect this
phys-q12-motor-torque Section II 3–5 marks mid range consensus 0.55

$F = BIl$ on a conductor or loop side, then torque on a current loop via $\tau = nIAB\sin\theta$ (or $F$ times perpendicular distance): compute the maximum torque and the torque at a stated angle, describe the variation through a half-rotation (max → zero at 90° of turn → max), and name changes that would double it. The 2025 Q23 trap — the area formula applied when only part of the loop lies in the field — is the flagged discriminator.

Each model's own prediction
  • Claude Fable 5: $F = BIl$ then torque, variation through a rotation, 3–5 marks (p 0.55)
  • Claude Opus 5: maximum torque then magnitude through a half rotation (p 0.45)
  • Gemini 3.1 Pro: multi-turn coil, maximum and 45° torque (p 0.75)
  • Grok 4.6: $\tau$ at the position shown, 180° description, two ways to increase it (p 0.48)
  • DeepSeek V4: maximum torque then back-emf effect on current and torque (p 0.50)

Marker-feedback lineage: Marking feedback 2021 Q21(b), 2024 Q21(b), 2025 Q23 (rotation-angle confusion, cm conversion, area-formula misuse)

In the practice paper: Q22

5 of 6 models expect this — consensus probability 0.54 Ideal transformer chained to transmission-line loss 5 of 6 models expect this
phys-q13-transformer-transmission Section II 3–6 marks mid range consensus 0.54

Turns-ratio calculation chained to $P = I^2R$ loss in a resistive transmission line at two voltages, closing with why high-voltage transmission wins and one efficiency feature (laminations against eddy currents, flux linkage, low-resistance wire). Three models add a switched or paralleled second load, where conservation of energy must carry the change back to the primary current — the exact step 2023 Q28(b) feedback says students omit.

Each model's own prediction
  • Claude Fable 5: turns ratio, line loss at two voltages, one efficiency feature (p 0.50)
  • Claude Opus 5: switched second load, primary current explained via energy conservation (p 0.55)
  • Gemini 3.1 Pro: system efficiency with losses located and named (p 0.70)
  • GPT-5.6 Sol: changing load, primary response, quantified line-loss saving (p 0.42)
  • Grok 4.6: 240 V, two globes and a switch, $P_\text{in} = P_\text{out}$ reasoning (p 0.52)

Marker-feedback lineage: Marking feedback 2022 Q22, 2023 Q28(b), 2025 Q22 (energy-conservation reasoning and generic transmission answers)

In the practice paper: Q30

5 of 6 models expect this — consensus probability 0.54 Crossed E and B fields: undeflected passage, then remove one field 5 of 6 models expect this
phys-q14-crossed-fields Section II 2–6 marks discriminator consensus 0.54

A charge passes undeflected through perpendicular electric and magnetic fields ($qE = qvB$): find the speed, a field strength or the required field directions, then predict the path when one field is switched off or the speed changes — the magnetic force scales with v, the electric force does not. MC-tested in 2020, 2022 and 2024 but never yet a full written item, which is the panel's case for promotion; opus and grok hedge it as MC again.

Each model's own prediction
  • Claude Fable 5: written velocity-selector promotion, path when a field is off (p 0.40)
  • Claude Opus 5: MC: velocity selector or E-vs-B trajectory comparison (p 0.70)
  • Gemini 3.1 Pro: find E given B and v for undeflected passage, 2–3 marks (p 0.70)
  • GPT-5.6 Sol: undeflected speed, field directions, then radius after E removed (p 0.51)
  • Grok 4.6: MC: second particle at 2v no longer balances (p 0.38)

Marker-feedback lineage: Marking feedback 2019, 2020 Q34, 2022, 2023 (field-type confusion and deflection-direction errors)

In the practice paper: Q28

Watch list — worth having ready (14)

  • Routine Faraday emf calculation from a B–t graph or withdrawn coil (fable 0.70, opus 0.60, gpt-5.6-sol 0.48) — squeezed out of the cluster list only by the cap; near-certain as a small item or MC

  • Satellite orbit-comparison MC: period, speed, $K$, $U$ and $E$ at two radii, with the less-negative-$U$ trap (opus 0.85, gpt-5.6-sol 0.55, grok 0.50)

  • Half-life read off a decay graph, then $\lambda = \ln 2 / t_{1/2}$ and $N = N_0 e^{-\lambda t}$ (fable 0.60, gpt-5.6-sol 0.52, opus 0.50, grok 0.50) — written decay work idle since 2022 Q24

  • Orbital energy transfer: $\Delta K$, $\Delta U$, $\Delta E$ between orbits and bound-or-escape judgements (grok 0.58, fable 0.50, opus 0.50, gpt-5.6-sol 0.32) — the 2025 Q36 lineage, argued down by some models precisely because 2025 just used it

  • Parallel-plate deflection chain: accelerate, deflect, project to a screen (opus 0.55, fable 0.50, deepseek 0.45) — the 2024 Q28 template

  • Rydberg/Balmer hydrogen-transition calculation, rested in 2025 (fable 0.60, opus 0.45, grok 0.45)

  • Blackbody-curve sketch or Wien routine calculation with the K/nm unit trap (deepseek 0.55, fable 0.50, grok 0.50)

  • Geiger–Marsden / atomic-model evidence, shrunk from the 9-mark essay to 3–5 marks or MC (fable 0.55 as extended response, deepseek 0.50, opus 0.45, grok 0.40)

  • Back emf written return after 2025 tested it only as MC Q11 (grok 0.50, deepseek 0.50, fable 0.45, gpt-5.6-sol 0.41)

  • Maxwell's EM prediction related to an oscillating-charge diagram — five models (0.40–0.47, mean 0.43) call a written return after a seven-year gap since 2019 Q25

  • Parallel-wires Newton's-third-law claim evaluation (opus 0.50, deepseek 0.45, fable 0.40, gpt-5.6-sol 0.38, grok 0.28) — grok argues 2024 Q29(b) spent the trap

  • AC generator output graph after speed is doubled — gemini's top-of-panel call (85% chance) with deepseek's MC variant (55% chance); the other four leave it inside Faraday items

  • Relativity contraction: gpt-5.6-sol's bold call (44% chance) that relativity drops to MC only after 2025's 8-marker; it is the panel's sole rester of the topic

  • Deep matter (Standard Model, accelerators) is the panel's likely-rested call: 4 of 6 models rest it and consensus P(examined) 0.669 / P(substantial) 0.265 are the lowest of all 12 topics

Where to spend your study time

How likely each topic is to appear this year.

Mod 6: EM induction, transformers, motors/generators 96% likely

Chance of a big question (4+ marks) here: 81%

Question types predicted here extended response ×6 short answer ×5 multiple choice ×2 stimulus based ×2

What each model expects

  • DeepSeek V4: A 5-mark Section II item combining Faraday's law calculation from a graph and an explanation of energy dissipation in an eddy-current braking system.
  • Claude Fable 5: A 5-8 mark Lenz's-law device analysis with a measured quantity to interpret sits late in Section II, extending the 2022-2025 run (exercise bike, roller-coaster brake, spinning can, pipe on a balance).
  • Gemini 3.1 Pro: A multi-part question requiring quantitative analysis of Faraday's Law from a flux or voltage graph, followed by a qualitative explanation of Lenz's Law.
  • GPT-5.6 Sol: A piecewise flux-time stimulus will combine Faraday’s Law, Lenz’s Law and the distinction between emf, current and force.
  • Grok 4.6: Section II will carry a 5–8 mark induction item — sliding rod, transformer conservation, or eddy-current energy — as the likely quantitative discriminator once gravity’s 8-mark slot is rested.
  • Claude Opus 5: For the sixth consecutive year the paper contains a magnetic-braking or eddy-current-damping scenario in an unfamiliar device, worth 4-7 marks and marked chiefly on the energy account.
Mod 5: Projectile and circular motion 95% likely

Chance of a big question (4+ marks) here: 76%

Question types predicted here extended response ×6 short answer ×6 multiple choice ×3

What each model expects

  • DeepSeek V4: A projectile question with non-zero vertical displacement between launch and landing, requiring solution of a quadratic for time of flight.
  • Claude Fable 5: The written projectile question again lands at a height different from launch, demanding a two-stage time-of-flight, and circular motion also carries a written force-analysis item.
  • Gemini 3.1 Pro: A classic 4-5 mark projectile motion problem with an asymmetrical trajectory (unequal launch/land heights) testing velocity resolution.
  • GPT-5.6 Sol: A late Section II mechanics problem will connect tangential launch from circular motion to an unequal-height projectile calculation.
  • Grok 4.6: Section II will include a 4–6 mark multi-step projectile with a non-zero net vertical displacement, not a simple symmetric parabola.
  • Claude Opus 5: A single question chains circular motion into projectile motion, taking the launch velocity from a rotation (period or revolutions per second) and asking for range or landing position.
Mod 7: Wave/particle behaviour of light, photoelectric 94% likely

Chance of a big question (4+ marks) here: 72%

Question types predicted here extended response ×5 short answer ×5 multiple choice ×3 practical analysis ×2

What each model expects

  • DeepSeek V4: A graph-based photoelectric effect question requiring determination of Planck's constant from data and explanation of particle model features.
  • Claude Fable 5: The wave-model side carries the written weight in 2026 - a quantitative double-slit-with-light or first written Malus calculation - while the photoelectric effect drops to a short item and MC after dominating 2025.
  • Gemini 3.1 Pro: A quantitative diffraction or double-slit question requiring calculation of fringe spacing, paired with a qualitative wave model explanation.
  • GPT-5.6 Sol: The substantial light question will favour quantitative interference or polarisation rather than another full photoelectric graph.
  • Grok 4.6: A 4–6 mark photoelectric or double-slit calculation in Section II, not a repeat of 2025’s stopping-voltage graph.
  • Claude Opus 5: Both models of light are examined in the same paper and probably in one question, with the wave evidence carrying a numerical fringe or grating calculation and the quantum evidence carrying a work-function calculation.
Mod 6: Charged particles in fields 93% likely

Chance of a big question (4+ marks) here: 66%

Question types predicted here short answer ×8 extended response ×4 multiple choice ×3

What each model expects

  • DeepSeek V4: A combined E-field and B-field problem comparing trajectories of electron and proton, likely 5 marks in Section II.
  • Claude Fable 5: The charged-particle question shifts from 2024's plate-deflection chain to magnetic-field circular motion or a first-ever written velocity-selector scenario.
  • Gemini 3.1 Pro: A heavy 5-7 mark question bridging electric field acceleration and magnetic field deflection (mass spectrometer principles).
  • GPT-5.6 Sol: A combined-field velocity-selector problem will chain force balance to circular motion and require an explicit magnetic-field direction.
  • Grok 4.6: A 3–4 mark B-field trajectory comparison (proton vs alpha) plus an MC on whether magnetic fields change kinetic energy.
  • Claude Opus 5: The paper again blends charged-particle deflection with projectile-style kinematics inside one multi-part question, mapped to both Module 5 and Module 6.
Mod 8: Origins of elements, stars, HR diagram 91% likely

Chance of a big question (4+ marks) here: 61%

Question types predicted here extended response ×5 multiple choice ×3 short answer ×3 stimulus based ×3

What each model expects

  • DeepSeek V4: An HR diagram interpretation question asking for fusion stage and evolutionary state of a specified star, worth 3–4 marks.
  • Claude Fable 5: An HR-diagram written question returns after two years at MC level, paired with a stellar-fusion identification part.
  • Gemini 3.1 Pro: A 4-5 mark question demanding the interpretation of an HR diagram to explain main sequence stellar evolution and nucleosynthesis.
  • GPT-5.6 Sol: A 4–6 mark Section II H–R diagram question will require evolutionary and fusion inferences rather than merely reading temperature and luminosity.
  • Grok 4.6: A 4–5 mark HR-diagram comparison in Section II, after 2025 reduced origins to a single MC.
  • Claude Opus 5: Origins returns to substantial weight after collapsing to one mark in 2025, most probably as an H-R diagram question of three to five marks naming the fusion process in each star's core.
Mod 8: Quantum models, radioactivity, fission/fusion 91% likely

Chance of a big question (4+ marks) here: 65%

Question types predicted here short answer ×8 extended response ×5 stimulus based ×1

What each model expects

  • DeepSeek V4: A binding energy and mass defect calculation linked to fission, with an explanation contrasting fission and fusion stability.
  • Claude Fable 5: A written mass-defect energy calculation returns after resting in 2025 (which kept the nucleus qualitative in Q21), probably chained with momentum or binding-energy reasoning as its discriminator.
  • Gemini 3.1 Pro: A quantitative calculation of mass defect and binding energy from a given nuclear decay equation.
  • GPT-5.6 Sol: Radioactive-decay modelling will return as a multi-part Section II calculation, with fission-chain control the leading extended-response coverage-gap candidate.
  • Grok 4.6: A mass-defect energy calculation will sit in Section II; the high-end version adds conservation of momentum to split KE between alpha and daughter.
  • Claude Opus 5: A mass-defect energy calculation returns after being absent from Section II in 2025, most likely inside a multi-part nuclear question that then asks why one product carries most of the kinetic energy.
Mod 5: Gravitation and orbits 90% likely

Chance of a big question (4+ marks) here: 64%

Question types predicted here multiple choice ×5 short answer ×5 extended response ×4

What each model expects

  • DeepSeek V4: A conservation-of-energy and angular momentum problem in an elliptical orbit, likely in Section II with 5–6 marks, including a derivation step.
  • Claude Fable 5: A 'show that' derivation (Kepler III or escape velocity) feeds a data-based calculation, mirroring the 2024 Q25 and 2025 Q24 structure, rather than another 8-mark orbital capstone.
  • Gemini 3.1 Pro: An orbit energy calculation testing gravitational potential energy changes between two orbital altitudes.
  • GPT-5.6 Sol: Gravity will be reduced mainly to orbital-property comparison, with no second consecutive eight-mark satellite synthesis.
  • Grok 4.6: Gravity appears, but not as the 8-mark closer: expect a 4–6 mark orbit-energy calculation plus one MC, after 2025 Q36’s explosion analysis.
  • Claude Opus 5: Gravitation carries the largest single Section II question or the second-largest, framed as an orbit change or explosion analysed by conservation of energy rather than by force alone.
Mod 7: Maxwell, EM spectrum, spectroscopy 87% likely

Chance of a big question (4+ marks) here: 54%

Question types predicted here short answer ×8 extended response ×3 practical analysis ×1 stimulus based ×1

What each model expects

  • DeepSeek V4: A spectroscopy question combining Wien's law and Doppler shift to determine star properties, with calculation.
  • Claude Fable 5: A written spectra-of-stars question returns after the 2025 rest, and Maxwell's predictions are the highest-value dust-covered dot point in Module 7.
  • Gemini 3.1 Pro: An analysis of stellar spectra to deduce temperature (Wien's Law) and radial velocity (Doppler shift).
  • GPT-5.6 Sol: Maxwell’s predictions will move from the 2025 multiple-choice item into a labelled-model Section II explanation.
  • Grok 4.6: A Wien calculation or black-body sketch, not another Roemer/Io speed-of-light reconstruction (2025 Q34).
  • Claude Opus 5: A spectroscopy question requires three distinct inferences from one supplied spectrum, with the third inference (rotation, motion or density) carrying the discriminating mark.
Mod 6: Motor effect, torque 87% likely

Chance of a big question (4+ marks) here: 48%

Question types predicted here short answer ×6 multiple choice ×3 extended response ×3 stimulus based ×1

What each model expects

  • DeepSeek V4: A 3-mark motor question linking torque calculation to back emf, likely in Section II.
  • Claude Fable 5: A 3-5 mark motor-effect calculation appears, and the Newton's-third-law parallel-wires misconception from 2024 Q29 resurfaces as an MC distractor or short evaluate item.
  • Gemini 3.1 Pro: A straightforward application of Ampere's force law between parallel wires or torque on a coil, testing angular dependence.
  • GPT-5.6 Sol: Back-emf reasoning will be tested through a changed-load graph, while direct torque calculation is likely confined to multiple choice or a short part.
  • Grok 4.6: A 4–6 mark torque calculation on a coil plus a back-emf current MC; not another 4-mark Newton-third rods evaluation.
  • Claude Opus 5: The motor effect is examined through force between parallel conductors rather than a single wire, with the discriminator being Newton's third law applied to unequal currents.
Mod 7: Special relativity 84% likely may rest

Chance of a big question (4+ marks) here: 49%

Question types predicted here short answer ×6 extended response ×2 multiple choice ×2

What each model expects

  • DeepSeek V4: A time dilation calculation tied to a cosmic muon scenario, with an explanation part linking length contraction and time dilation from different frames.
  • Claude Fable 5: Relativity retreats from the 2025 8-mark essay to a 3-4 mark calculation plus one or two MC items; muon evidence stays at MC level after the 2024 written treatment.
  • Gemini 3.1 Pro: A 4-5 mark calculation on muon decay testing both time dilation and length contraction from two reference frames.
  • Grok 4.6: Relativity shrinks to a 3-mark time-dilation calculation plus an invariant-c MC, after 2025 Q32’s 8-mark essay.
  • Claude Opus 5: Relativity is present but compact in 2026, a calculation of three to five marks plus a multiple choice, rather than the eight-mark synthesis it carried in 2025.
Mod 8: Structure and models of the atom 73% likely may rest

Chance of a big question (4+ marks) here: 52%

Question types predicted here extended response ×4 short answer ×4 multiple choice ×2

What each model expects

  • DeepSeek V4: A 5-mark comparison of three atomic models focusing on electron nature, with a diagram, targeting the deep understanding demanded in feedback.
  • Claude Fable 5: The M8 extended response rotates back to atomic-structure evidence (Geiger-Marsden/Chadwick era) after 2025's cathode-ray and Standard Model focus, and a Rydberg-equation calculation returns.
  • Grok 4.6: A 3–4 mark named-experiment item (Thomson/Millikan or de Broglie vs Bohr), not another 8–9 mark ‘development of the atom’ essay.
  • Claude Opus 5: A four-to-five mark experiment-to-model question appears again, and Geiger-Marsden is the most likely subject after cathode rays in 2025 and Chadwick in 2024.
Mod 8: Standard model, particle accelerators 67% likely may rest

Chance of a big question (4+ marks) here: 27%

Question types predicted here short answer ×2 multiple choice ×1

What each model expects

  • DeepSeek V4: A quark-level beta decay question in MC, plus a short-answer linking particle accelerator energy to wavelength for probing structure.
  • GPT-5.6 Sol: Deep matter will contract to a reaction-validity or quark-conservation item rather than another broad Standard Model essay.

How likely each topic is to appear. Open a topic for the question types to practise there.

Practise with the paper

Physics practice paper

100 marks · 36 questions

Every question is traceable to the consensus prediction behind it — open the web version and each question carries a “why this question” link into the evidence. All questions are original Intuition compositions in NESA style.

Read it on the web Download PDF Solutions (PDF)

Exam Success Program

Book our ESPs now — online or in person

A paper on your desk is one thing — the clock running is another. Sit Physics practice exams under real conditions and get them marked with real feedback. Exam times run seven days a week, right up until the HSC begins.

Book your ESP →

Places are still open for 2026

Intu AI

One paper isn't enough? Generate more

Intu AI builds unlimited practice questions for Physics in these styles, marks your working, and explains what you missed — aligned to your syllabus.

Practise with Intu AI →

Share & save

Practice paper (PDF)

Every style card and paper question has its own link — hover any card and use its copy-link icon to share exactly the thing you mean. Printing this page gives a clean copy too.

Published Aug 2026, before the exams. In November 2026 we score these predictions publicly against the real paper — per-model calibration and question-level hit rates, the same harness as the 2025 backtest. How we did it.