On 30 April, Summit Therapeutics (SMMT) announced that the data monitoring committee had recommended that the HARMONi-3 squamous cohort continue after its interim PFS analysis. Summit has not said the interim bar was missed. However, the protocol has no futility look that “continue” could refer to, management did not dispute a question framed as “missing that interim” on the Q2 call, and a crossing on a primary endpoint would have been material. We therefore treat the miss as a strong inference. Summit expects the events for the final PFS analysis in 2H26, and the question is how much information the miss carries.

That depends on one number Summit has not disclosed: t, the share of final PFS events available at the interim. An early interim, at ~60% of events, makes the miss a modest update, and the final keeps a 63% chance of success on our model. A later one, at ~80%, makes it a heavy update and cuts that chance to 36%. Summit added the interim by amendment and timed it by calendar, weeks after enrollment closed, which points to the early end. We center t at ~0.65 (calendar range 0.60 to 0.70), which puts the chance of final success at ~55% to 60% (57% on central assumptions).

Ivonescimab is a PD-1/VEGF bispecific that re-rated Summit’s shares in May 2024, when Akeso’s China-only HARMONi-2 trial showed it beat pembrolizumab (Keytruda) on PFS as monotherapy in PD-L1-positive NSCLC (HR 0.51); survival has since followed. HARMONi-3 is the first global test of ivonescimab plus chemotherapy against pembrolizumab plus chemotherapy in 1L NSCLC, the setting behind a large proportion of Keytruda’s revenues. Pembrolizumab plus chemotherapy is the established option, yet five-year survival on it in squamous disease is below 20% (KEYNOTE-407). A regimen that adds a durable benefit with acceptable toxicity, shown first on PFS and then on OS, would be practice-changing.

Summit holds the rights in the Americas, EMEA and Japan, and has a fully diluted market capitalization of ~$15.0B at the 14 Oct close of $15.13. FDA’s decision on HARMONi, its first US filing (EGFR-mutated NSCLC), is due on 14 Nov 2026.

The debate that has followed the stock is whether Chinese data translates to Western patients. Akeso’s positive Ph3s were all run in China, and population differences, from smoking patterns to tumor biology, are possible but unestablished sources of a gap. The one multiregional readout, HARMONi, was PFS-positive but missed its primary OS endpoint, and its Western PFS HR (0.67) was weaker than its Asian one (0.55). Bulls see an effect that should carry; skeptics see a data set that may not. The interim miss sharpens that debate without settling it: if the interim was weighted to Chinese patients and the final adds more Western ones, the final print could be worse than the interim implies.

This note reconstructs the undisclosed HARMONi-3 design from KEYNOTE-407, estimates t, builds two priors for the true effect, conditions them on the miss and projects the final print (Sections 2 to 5). It then reads each possible print three ways, statistically, clinically and for the investment case (Section 6), because a print that clears the bar can still fall short of what physicians or investors need.

1. What the final has to show

HARMONi-3 compares ivonescimab plus chemotherapy with pembrolizumab plus chemotherapy in 1L metastatic NSCLC, in squamous and non-squamous cohorts. The squamous cohort (~600 patients) reads out first, with an early OS look alongside the final PFS analysis. Summit added the interim PFS analysis in Feb 2026 by amending the statistical analysis plan, and timed it for Q2, weeks after squamous enrollment closed in Q1, to open an earlier conversation with regulators. Summit said the interim spent minimal alpha and set a “meaningfully higher bar” than the final.

Treating the miss as a strong inference, the question is what it says about the true effect, and what the final print then has to show to matter. A successful final has to pass three tests, and a print can pass one and fail the next:

  1. Statistical. Does the observed HR clear the final bar?
  2. Clinical. Is the effect large enough to change practice for physicians and regulators?
  3. Investment. Does the result change what the franchise is worth?

On our model a statistical pass is far more likely than a clear win (Table 5). A print of ~0.78 would clear the bar and still sit well short of HARMONi-6’s 0.60. Sections 2 and 3 set up the statistical and clinical tests, and Section 6 shows where the three readings diverge.

Our call. We put the chance that the final PFS analysis succeeds at ~55% to 60% (57% on central assumptions), down from ~84% before the interim. Our central estimate of the print is an HR of ~0.78, just inside a final bar of ~0.79, with a 16% chance of a clear win (below 0.70). Two risks we could not model, narrowing after month 9 and a more Western final, both push the print higher, so we treat 57% as a central figure with downside skew. The call assumes the interim missed its efficacy bar; if it was not a formal efficacy miss, the ~84% pre-interim figure is the relevant frame.

We have no valuation bridge and no joint PFS/OS probability model, so the note describes what each outcome implies and does not claim to know what HR is priced in.

Reported, inferred and unknown. Reported: the design, cohort sizes, timing, independent-review PFS and the HARMONi, HARMONi-2 and HARMONi-6 results. Inferred: the miss, t, the event targets, the alpha split, the China share and the expected HR. Unknown until the print: the interim cut date, actual squamous enrollment and the regional mix at each look.

2. The design Summit has not disclosed

The statistical test needs a bar, and Summit has not disclosed the alpha split or the event targets that set it. We rebuilt the design on the closest precedent, KEYNOTE-407, the ~560-patient squamous trial that made pembrolizumab plus chemotherapy the standard. Its final protocol (Nov 2017) set a one-sided PFS alpha of 0.010, an interim at ~332 events (80%) and a final at ~415 events, for 90% power at HR 0.70. Those choices produced observed-HR bars of 0.747 at the interim and 0.792 at the final.

We rebuilt HARMONi-3 on this template with the Schoenfeld formula. The method reproduces KEYNOTE-189’s 465 events (protocol target 468) and HARMONi-6’s 293 events (protocol target 297). At t of 0.80 and 415 events, it also returns KEYNOTE-407’s bars to the third decimal (Figure 1; Table A1 in the appendix). Our locked case uses independent-review PFS (EU protocol v6.0 and a Mar 2026 fact sheet override the registry’s “investigator” label), PFS alpha 0.010 and 415 final events.

The event target is inferred too. A 90%-power design at HR 0.70 and alpha 0.010 needs ~410 to 415 events, KEYNOTE-407’s own number. The disclosed calendar fits that target, though it cannot pin it down. We ran a patient-level simulation of 400,000 trials (415 events, enrollment back-loaded to mid-Mar 2026, KEYNOTE-407’s mature control median of 8.0 months). In the trials that missed the interim at t of 0.65, the median interim cut fell in mid-to-late Mar, in time for review by 30 Apr. The median final cut fell in late Oct to late Nov, in the back half of 2H26. A 370-event design (86.3% power, HARMONi-6’s figure) would have reached both looks months earlier unless the control median were ~10 months. We therefore model 415 and show 370 as the alternative. At 370, the calendar implies a later interim (t of ~0.72 to 0.78) and blended odds of ~47% to ~39%.

At 415 events the final bar is an observed HR of ~0.794 (~0.783 at 370), so a printed 0.79 passes in our locked case and would fail at 370. The test itself is a stratified log-rank p-value, so the HR bar is an approximate guide. The interim bar runs from ~0.67 to ~0.75 across t of 0.60 to 0.80, while the final bar moves by less than 0.005, so the interim cost the final almost no alpha.

The hinge is t. The precedents placed their interims by event count: KEYNOTE-407 and KEYNOTE-189 at ~80% of final events, HARMONi-6 at a planned 70% (run at 74%). HARMONi-3’s interim was added by amendment and timed by calendar, weeks after the last squamous patient enrolled. Management said many patients had “not a lot of follow-up time” at the interim, so it held fewer events than an event-driven look would have, although more than 80% of the ~600 patients were enrolled by Oct 2025. On a 415-event design, our accrual curve makes an interim cut between mid-Mar and mid-Apr consistent with t of ~0.60 to 0.70; a more front-loaded enrollment would put the same window nearer 0.70 to 0.75, which is why t is the main risk to the call (Section 7). A faster control arm (KEYNOTE-407’s early 6.4-month median) would put t at ~0.76. It would also have put the required final events around Aug, and on the 23 Jul call management said reaching them was “not next month”. We center t at ~0.65, treat 0.60 to 0.70 as the calendar range and keep 0.80 as the upper bound.

The named alternative mirrors HARMONi-6’s hierarchy: PFS alpha 0.025 with ~300 events and a final bar of ~0.79 to 0.80. It changes the decoder more than the post-miss odds (Section 5).

That defines the statistical test. The final needs an observed HR of ~0.79 or better, while the interim, run early in follow-up, probably needed ~0.70. The harder question is what the true effect is likely to be.

3. The China question

The clinical test depends on the true effect, and almost all the evidence on it is Chinese. HARMONi-3 is the first trial of ivonescimab plus chemotherapy against pembrolizumab plus chemotherapy in a global population, and neither close anchor tests that regimen. HARMONi-2 compared ivonescimab with pembrolizumab, both as monotherapy, in PD-L1-positive patients (TPS ≥1%). HARMONi-6 compared it against tislelizumab plus chemotherapy. HARMONi-3 enrolls irrespective of PD-L1. The only multiregional test, HARMONi, missed its primary OS endpoint (HR 0.79, p=0.057).

We estimate China at 39% of the squamous cohort (range 36% to 42%). The anchor is a UBS-reproduced screenshot of the China Drug Trials registry entry (CTR20232457, dated 15 Dec 2024), which we could not independently verify: 383 of 1,080 planned patients in China (~35%), with 175 already enrolled. We set the squamous share above 35% on our judgment that squamous enrollment is more China-weighted. ClinicalTrials.gov lists ~1,600 estimated patients and no regional counts, and Summit has not given the mix at any look.

HARMONi-6 was also uneven inside the trial. PFS HR was 0.40 (0.26 to 0.59) under age 65 and 0.88 (0.61 to 1.27) at 65 or older, with non-overlapping intervals but no formal interaction test (an unstratified subgroup cut), and 0.71 (0.37 to 1.33) at PD-L1 ≥50%. Median age (64 vs 65 in KEYNOTE-407) does not separate the trials. Much of the 0.60 headline, however, comes from one subgroup, so the anchor is less portable, and we shrink it before using it (Section 4).

Geography enters the model through a Western shift. HARMONi’s total PFS analysis (Apr 2025 cut, 438 patients) gave an HR of 0.67 (0.45 to 1.00) in North America and Europe and 0.55 (0.43 to 0.71) in Asia, a gap of 0.20 on the log scale. That gap comes from a different setting (EGFR-mutated disease after a TKI, placebo control), and the Western interval reaches 1.0. We therefore apply a shift of 0.12, about 60% of the gap, to the ~61% non-China share.

The size of that shift matters more than the China share. Moving the share five points either way, wider than our range, moves the median true HR by less than 0.01 and post-miss success by 2pp or less. Applying HARMONi’s full regional gap lowers the central 57% to 52%.

Not modeled: composition across the looks. A China-heavy interim followed by a more Western final would make the final HR worse than the interim implies, and would push post-miss odds lower.

Section 4 turns the HARMONi-6 and HARMONi-2 anchors and this Western shift into two priors on the true effect.

4. Two priors on the true effect

Section 2 set the bar and Section 3 the geography. This section turns them into two priors on the true global PFS HR. Each prior describes the effect we would expect before any interim information, the yardstick against which the miss is judged; neither is a forecast of the final print.

Prior A is the skeptical case. It anchors on HARMONi-6, the squamous-only Ph3 against tislelizumab plus chemotherapy (any PD-L1), where the independent-review PFS HR was 0.60 (95% CI 0.46 to 0.78; medians 11.1 vs 6.9 months). We shrink that result for the winner’s curse (prior SD 0.35), which gives 0.641, and then apply the Western shift of 0.12 on the log scale to the ~61% non-China share (Section 3). The result is a median true HR of 0.690, with a 54% probability that the true HR is below 0.70.

Prior B is the optimistic case. It starts from HARMONi-2’s unstratified squamous PFS HR of 0.50 (0.33 to 0.76; overall stratified 0.51), a descriptive subgroup comparing ivonescimab with pembrolizumab, both as monotherapy, in PD-L1-positive patients. We keep 74% of that effect, which gives a median of 0.595. Its log-SD of 0.20 is wider than Prior A’s, because the subgroup’s own sampling error, scaled by the 74% retention, is already ~0.16 before any bridging uncertainty. HARMONi-2 has since reported an OS HR of 0.73 (p=0.009) overall and 0.65 in squamous patients (n=181), though the squamous figure is also descriptive.

We blend the two at equal weight before the interim and reweight by each prior’s miss probability afterwards. The equal weight is our judgment rather than a fitted parameter, and Section 5 shows how little the answer moves when it changes. Table 1 sets the two priors side by side.

In months, the priors depend on which KEYNOTE-407 control figure applies. Pembrolizumab plus chemotherapy had a median PFS of 6.4 months at the 7.8-month interim and 8.0 months at the 14.3-month final, and the calendar check favors the mature 8.0-month figure. On that range, Prior A implies ~9.3 to 11.6 months for ivonescimab, a gain of ~2.9 to 3.6 months, while a print at the final bar of ~0.79 implies a gain of only ~1.7 to 2.1 months. That gap is the clinical test expressed in months. The figures assume exponential survival, and HARMONi-6 ran against a different control, so its 11.1 vs 6.9 months compares with HARMONi-3 only as a ratio.

Before the interim, both priors put the true HR comfortably inside the final bar of ~0.79: Prior A gives a 78% chance of success, Prior B 90% and the blend ~84%. The next section asks how far a miss should move that.

5. What the miss does

A miss rules out the interim results that would have cleared ~0.70, and with them much of the probability that the drug is strong. We track four objects through the update: the prior true effect, the posterior true effect after a miss, the predicted final observed HR, and the chance that the final crosses its bar. Table 2 gives the results at PFS alpha 0.010 and 415 events.

On Prior A, the miss moves the median true HR from 0.690 to 0.759, and it shifts weight toward the skeptical prior, from 50% to 66% on Prior A. Because Summit timed the interim by calendar, weeks after enrollment closed, a miss at our central t of ~0.65 says less than a precedent-style look at 80% of events would have.

The predicted final print sits ~0.02 to 0.04 above the posterior true effect, because the final analysis contains the interim data that already fell short. At t of 0.65 the median print is 0.782 on Prior A and 0.777 on the blend, inside the bar of ~0.794 (Figure 2); on Prior A it reaches the bar by t of 0.70. The chance of a print below 0.70 is 14% on Prior A and 16% on the blend.

Together these give a 55% chance that the final crosses on Prior A and 57% on the blend (51% to 63% across the calendar range), which we publish as ~55% to 60%. Against ~84% before the interim, the miss costs ~27pp. At 370 events the chance of crossing differs by 1pp to 2pp, though the print bands move more: on Prior A at t of 0.65, the chance of a print below 0.70 rises from 14% to 18%.

What moves the answer. t is the strong lever. Moving it from 0.60 to 0.80 cuts post-miss success from 63% to 36% on the blend (Figure A1 in the appendix), while the China share moves it by 2pp or less (Section 3). The alpha split matters little as long as spending is O’Brien-Fleming-like, and moving the pre-interim weight on Prior B to 25% or 75% gives 56% or 59%. The spending function is undisclosed and does matter. A Hwang-Shih-DeCani function that spends about as little at the interim (gamma −6) gives 59% on the blend, and one that spends more (gamma −4) gives 53%. Pocock spending would give 32%, but it puts ~75% of the alpha at the interim, which does not fit Summit’s “minimal alpha” description.

The named alternative from Section 2, PFS alpha 0.025 with ~300 events, gives similar odds: 59%, 47% and 33% on Prior A at t of 0.6, 0.7 and 0.8 (62%, 49% and 35% on the blend), with median prints of 0.772, 0.800 and 0.827. Its bar is ~0.79 to 0.80, so the same print is judged more leniently.

Table 3 shows how far the central 57% moves when one input changes at a time.

Most single changes keep the answer between ~50% and ~60%. The ones that break below 50% are a larger Western penalty, a weaker retained Prior B, a worsening effect between looks and Pocock-type spending. The worsening row matters most for the call: a drift of 0.04 on the log scale is small, and it is the direction the HARMONi-6 narrowing points.

6. Reading the print

6.1 The decoder: three readings of each print

The three tests from Section 1 return here. A print can clear the statistical bar and still disappoint physicians, and a clinically good print can still leave the investment case dependent on OS and safety. The decoder (Table 4) gives three readings of each possible final HR. At alpha 0.010 the statistical bar is ~0.79 at 415 events (~0.78 at 370). The clinical bands (below 0.75 “good”, below 0.70 “clear win”) come from one physician on an investor-hosted call, so they are a single-source view. The older ~0.80 floor holds only at alpha 0.025.

The decoder does not say what the stock requires. A print in the 0.70 to 0.75 band is the clearest case where all three readings point the same way. The 0.75 to 0.79 prints are where the statistical and clinical readings diverge, and where the OS look will carry more weight.

Table 5 maps the model’s predicted final print onto the decoder bands at our central t of ~0.65.

Our central expectation sits just inside the bar: a statistically positive print of ~0.78, well short of HARMONi-6’s 0.60. In that zone the drug wins on statistics, and the clinical case depends on how far below 0.75 it lands. A print below 0.70 would fit an interim that came early, a skeptical prior that was too heavy, or simply good luck in the later data. A print above the bar would fit narrowing after month 9, a more Western final, or both.

6.2 Durability: will the benefit hold?

Durability is the main PFS risk. The PFS curves in HARMONi, HARMONi-A and HARMONi-6 separate early, so the late-benefit pattern from HARMONi OS does not apply. The live question is narrowing, and HARMONi-6 already shows it. From the published numbers at risk, the HR was ~0.58 over months 0 to 9 (189 events) and ~1.3 after month 9 (32 events, interval ~0.6 to 2.7). Most of the separation came in months 3 to 6, where the HR was ~0.41. The landmark rates tell the same story: 62% vs 43% at 9 months and 48% vs 35% at 12 months, which imply a cumulative HR of ~0.57 at 9 months and ~0.70 at 12. The late window is thin and could reflect censoring, so we read it as directional only. It matters because HARMONi-3’s final will carry more follow-up beyond month 9 than HARMONi-6’s interim did (median 10.3 months), which pushes the observed HR up relative to our constant-HR model.

Figure 3 shows the two simulated outcomes. Where the final crosses, the median gap is ~3.0 months (~11.1 vs ~8.1, in line with HARMONi-6 in absolute months). Where it fails, the gap is ~1.2 months (~9.5 vs ~8.3). The curves assume a constant HR, so they show no narrowing, which the HARMONi-6 evidence above says may happen after month 9.

The timing of the final carries a weak signal of its own. In the simulation, trials whose final crosses reach 415 events later (median cut late Nov 2026) than trials whose final fails (late Oct), because a weaker drug produces progressions faster. Events can also slow for reasons unrelated to the drug, and the spread is wide, so we read a later cut as only mildly favorable.

6.3 The OS bridge: what a PFS print does not settle

OS gets a bridge and no committed HR. The squamous cohort has four OS looks under the protocol: interim 1 at the Apr PFS interim, interim 2 with the PFS final in 2H26, interim 3 in 1H27, and a final analysis later. Management calls all of them formal analyses that spend alpha. The 2H26 look will be immature, so we read it for direction. Interim 3 is the first not tied to a PFS analysis, with median follow-up expected “in the low 20s” months, and the first that can carry real weight.

The OS looks also matter more for ivonescimab than usual. Summit’s Q2 10-Q records FDA’s position on HARMONi: a statistically significant OS benefit is necessary to support approval in that setting, and the PFS results may not be sufficient after OS missed at the primary analysis. That does not set the bar for HARMONi-3, where PFS and OS are dual primary endpoints, but it shows FDA will not treat a PFS win for this drug as enough on its own.

The protocol links the two endpoints in both directions. OS is tested at its initial alpha allocation, or at a higher level if PFS succeeds and passes its alpha on, so a PFS win at the final raises the OS alpha at the same look. Under the graphical procedure, an OS success at a later look passes alpha back to PFS, which is then retested at a higher level. A PFS miss at the 2H26 final is therefore not the last word, but we do not add this route to our odds, because it needs an OS win first, most likely no earlier than OS interim 3 in 1H27.

Early OS looks are hard to win. Under O’Brien-Fleming spending with the full one-sided 0.025, a look at half of the final deaths spends ~0.0015. With 200 deaths that needs an observed OS HR of ~0.66; at 40% of deaths, ~0.62. These bars are illustrative, since Summit has not disclosed deaths or information fractions.

What would worry us at OS interim 2 is a point estimate at or above 1.0; a reading between ~0.85 and 1.0 would be uncertain rather than negative at this follow-up. HARMONi’s Western OS HR moved from 0.98 to 0.84 to 0.76 across three cutoffs, at Western median follow-up of 9.2, 13.7 and 23.2 months (Figure A2 in the appendix). That shows how wide early uncertainty can be, but as a nominal 165-patient subgroup whose interval includes 1.0, it says little about whether an unfavorable early reading will improve. A look-by-look OS decoder would need deaths at each look, the information fraction, the alpha split and squamous enrollment, none of which Summit has disclosed. Table 6 instead translates PFS into OS with the ratio of log HRs from earlier trials. Every ratio puts OS closer to 1 than PFS, but the spread is wide: a PFS print of ~0.78 corresponds to an OS HR of ~0.82 to 0.89.

HARMONi is the cautionary case: an OS HR of 0.79 still missed the formal bar. HARMONi-6 is the clean version: OS HR 0.66 (0.50 to 0.87) at 204 deaths in the prespecified interim analysis, p=0.0017 against a 0.0049 boundary. Management calls an OS HR below 0.80 the “generally accepted clinical meaningfulness threshold” (Q2 call, 23 Jul).

7. What to watch, and what could break the call

Six items would change our view most after the print, in rough order of importance:

  1. The printed final HR, interval and event count. These fix the decoder row and resolve the 0.784 to 0.794 ambiguity.
  2. Any disclosure of the interim fraction or event count. Either would replace our t with a number and make most of Section 5 obsolete.
  3. PFS by region. This tests the Western shift directly.
  4. Curve shape after 9 to 12 months. Early separation that holds, or narrowing.
  5. When the final cut happens. A later cut is mildly consistent with a stronger effect; read it with care.
  6. The direction of OS interim 2. It comes with the PFS final and is immature, so we read it as direction only. A point estimate at or above 1.0 would worry us.

What could break the call

  • t is the main risk. Our calendar range is 0.60 to 0.70, and at its top end the odds are about a coin flip. Above ~0.71 the blended odds fall below a coin flip. Below 0.60 the miss carries less information and the post-miss odds move back toward the pre-interim figure; our tables do not go that low.
  • A 370-event final lowers the odds, because the calendar would then imply t of ~0.72 to 0.78 and the final bar would fall to ~0.78.
  • The Western PFS effect could be smaller than our shift. HARMONi’s full regional gap would lower the central 57% to 52% (Table 3).
  • Composition across the looks could be worse than we assume. A China-heavy interim followed by a more Western final is outside the model.
  • Narrowing after month 9 would push the final print above what a constant-HR model predicts, and a drift of 0.04 on the log scale alone takes the odds to 44%.
  • The control arm could do better in a Western-heavy population than the 6.4 to 8.0 months we calibrated from KEYNOTE-407. That would move every statement in months and the calendar mapping to t; the HR bars, decoder thresholds and Table 6 ratios would not move.
  • The event counts and the alpha split are inferred from KEYNOTE-407, and the actual protocol may differ.
  • We have no valuation bridge, so we cannot say which print the stock requires.

Context: other news flow

None of the three items below changes the squamous PFS reading, but each bears on how the print will be received.

FDA, 14 Nov. The HARMONi decision covers a different setting (2L, EGFR-mutated, post-TKI) and reads on regulators, not on HARMONi-3’s effect size. FDA’s stated OS requirement in that review is covered in Section 6.3. The closest precedent is sintilimab, where on 10 Feb 2022 the ODAC voted 14 to 1 that additional trials demonstrating US applicability should precede a final decision on a China-only package; HARMONi was multiregional, so the parallel is partial. On safety, HARMONi treatment-related serious adverse events were 28.0% vs 15.1%. In HARMONi-6 (ASCO 2026 update), treatment-related grade ≥3 events were 69.2% vs 58.9%, and grade ≥3 hemorrhage, counted among possibly VEGF-related events, was 2.6% vs 0.8%.

AstraZeneca. AstraZeneca invested $2.0B in Class A convertible preferred ($18.3561 per common-equivalent share, ~108.96M shares), signed 28 Sep and completed 5 Oct. It has no board seat, a one-year lock-up and a commitment to vote with Summit’s board. If the board actively considers a third-party acquisition proposal, AstraZeneca gets notice, 10 business days to make its own offer and diligence access, but no right of first refusal, matching right or veto (purchase agreement, Exhibit 10.1, §§5.14 to 5.18). That is a privileged seat in a sale process, short of control. With Summit’s $690.7M in cash, cash equivalents and short-term investments at 30 Jun, pro-forma liquidity is ~$2.5B or more. The financing lowers financing risk and says nothing about the data.

Competitor clocks. Among PD-(L)1xVEGF bispecifics in global 1L NSCLC Ph3s against pembrolizumab plus chemotherapy, Summit is ahead on the clock, which gives an earlier chance to establish evidence but no commercial lead. ROSETTA Lung-02 (BioNTech and Bristol Myers Squibb, pumitamig) lists primary completion in Jan 2029, and Symbiotic-Lung-01 (Pfizer, tivunatamig; 1,410 patients) in Feb 2029. The nearer readout is a different class: AstraZeneca’s AVANZAR (Datroway plus Imfinzi and chemotherapy, 1,350 patients, against pembrolizumab plus chemotherapy), where AstraZeneca guides to data in 2H26 against a registry completion date of 1 Nov 2027. Its primary PFS and OS analyses are non-squamous only (overall and TROP2 biomarker-positive), so a win would be indirect read-across to HARMONi-3 squamous.

Appendix

Sources