It is seven years since astronomers were first able to pinpoint the origin of a γ-ray burst (GRB) in a distant galaxy1,2,3. Since then, estimates of the energy associated with these titanic explosions have waxed and waned. Using only the brightness and distance of these objects, calculations had suggested that an amount of matter almost equal to the mass of the Sun was being transformed into γ-rays with near-perfect efficiency. But then evidence emerged for ‘beaming’ — the compression of the burst energy into narrow beams, or jets — and the energy estimates went down by a factor of about a hundred, to an amount equivalent to merely a few supernovae4. New evidence presented by Sazonov et al.5 and Soderberg et al.6 (starting on page 646 of this issue) suggests that the bursts observed so far are but the bright tip of the iceberg. It seems likely that many fainter explosions have gone unseen simply because our detectors weren't sensitive enough.

The first indication that GRBs might have a broad range of energies came in 1998 with the discovery7 of a nearby event, ‘only’ 130 million light years away (equivalent to a redshift, z, of 0.0085). Despite its proximity, GRB 980425 — named for the day on which it was observed — was not particularly bright. Its inferred total energy, emitted over 20 seconds, was only 7 × 1047 erg — about 10,000 times less than a typical GRB, but still 10 trillion times more than the Sun emits in the same time. The burst was also ‘soft’ compared with other GRBs: no prompt emission was seen8 above an energy of 300 kiloelectronvolts (keV); and it was accompanied by a very bright supernova9, SN 1998bw, whose extreme luminosity, velocity, and radio emission placed it in a class by itself. Because of its low energy, however, astronomers argued as to whether GRB 980425 really was associated with the nearby supernova (from which the redshift was calculated), or whether it just happened to be in the same general direction but much farther away. Others argued that, even if the GRB and supernova were connected, the GRB was an unusual one, in a class by itself, with little relation to the others.

Time and further observation have diminished those arguments. X-ray observations of SN 1998bw spanning several years10,11 show a smooth evolution in brightness that matches that of the GRB, from just one day after its appearance. Moreover, as inferred from radio and X-ray emission, the energy in the relativistic ejecta (the matter thrown out at more than 90% of the speed of light) was about 100 times more than that of the burst itself12, and hence comparable to that of other GRBs. For some reason, GRB 980425 put a huge amount of energy into making a supernova (1052 erg), a lot of energy into relativistic ejecta (1050 erg), but relatively little into γ-rays moving towards Earth (1048 erg). Why was its γ-ray emission so low? And, given its faintness, might there be many other similar events that have escaped detection?

Then came GRB 031203. This burst was discovered from space last December by the γ-ray telescope INTEGRAL; the details of its γ-ray emission are now reported by Sazonov et al.5 and Soderberg et al.6. Like GRB 980425, the new burst lasted about 20 seconds and was close to our Galaxy. Most GRBs have a redshift that is greater than one, but the host galaxy of GRB 031203 was at z = 0.1055 (ref. 13) — making it, at 1.6 billion light years, the next-closest event after GRB 980425. From the distance and observed flux, the inferred energy of GRB 031203 was 0.6–1.4 × 1050 erg, intermediate between GRB 980425 and more typical bursts. As the burst faded, a supernova was revealed14,15, SN 2003lw, with properties similar to SN 1998bw. It looked as if a twin to GRB 980425 had been found.

But there were differences. Although faint compared with most bursts, GRB 031203 was a hundred times more luminous than GRB 980425; and its spectrum is ‘hard’, more like that of a typical GRB. A spectrum is described as hard if the typical energy of its γ-rays is high. Many scientists had come to accept (if not fully understand) an empirical relation16, called the Amati relation, between hardness and a burst's total energy. But the hard spectrum of GRB 031203 promises to upset this apple-cart: the Amati relation would suggest a γ-ray energy of about 10 keV for GRB 031203; instead, INTEGRAL measured5 values in excess of 190 keV.

Another similarity between GRB 980425 and GRB 031203 was the energy they released in relativistic ejecta, inferred from their afterglows — the X-ray, radio and optical emission that follows the main burst for up to a few weeks, and now measured for GRB 031203 by Soderberg and colleagues6. For both GRBs, this energy was lower than average. Although the modelling of the process is uncertain, GRB 031203 may have been up to ten times less energetic6 than GRB 980425. The afterglow is created as the relativistic ejecta from the burst is slowed down by the surrounding medium. The emission from this more slowly moving matter is not nearly as concentrated into beams as the initial burst of γ-rays, so the energy in the afterglow is often regarded as a better measure of the total energy in the burst than the GRB itself (which is derived only from that part of the burst directed towards Earth). In fact, if the energies in relativistic ejecta and in the GRB itself are added together for GRB 980425 and for GRB 031203, these bursts are even more similar and distinctly less powerful than all other bursts analysed so far6.

So it seems that GRBs must now be considered a broad class of phenomena whose energy scale may vary by at least four orders of magnitude, even when they share such properties as spectrum, timescale, variability and the presence of a supernova. Perhaps this is not surprising. Models of GRBs offer considerable leeway in the partitioning of energy between supernova and relativistic ejecta, as it depends on such quantities as the rotation rate, mass and size of the parent star. However, given that most GRBs are beamed, their properties will inevitably vary depending upon the angle at which they are viewed. Sorting out both these effects — viewing angle and variable explosion energy — will require the study of many more bursts, especially the faint ones. Yet it is hard to detect events such as GRB 031203, which would have gone undiscovered if it had been just a few times more distant.

Fortunately, a new GRB mission, named Swift, will be launched by NASA this October. With its greater sensitivity, Swift should be able to detect many more faint bursts. Once it was thought that a more sensitive detector was not needed particularly urgently, because we could already see GRBs out over the entire observable Universe. But the data5,6 from GRB 031203 suggest that we may have been studying only the brightest ones.